Toshio Saito, Martin Scharlemann and Jennifer Schultens - Lecture Notes On Generalized Heegaard Splittings

a
r
X
i
v
:
m
a
t
h
/
0
5
0
4
1
6
7
v
1

[
m
a
t
h
.
G
T
]

8

A
p
r

2
0
0
5
LECTURE NOTES ON GENERALIZED HEEGAARD
SPLITTINGS
TOSHIO SAITO, MARTIN SCHARLEMANN AND JENNIFER SCHULTENS
1. Introduction
These notes grew out of a lecture series given at RIMS in the summer of 2001.
The authors were visiting RIMS in conjunction with the Research Project on Low-
Dimensional Topology in the Twenty-First Century. They had been invited by
Professor Tsuyoshi Kobayashi. The lecture series was rst suggested by Professor
Hitoshi Murakami.
The lecture series was aimed at a broad audience that included many graduate
students. Its purpose lay in familiarizing the audience with the basics of 3-manifold
theory and introducing some topics of current research. The rst portion of the
lecture series was devoted to standard topics in the theory of 3-manifolds. The
middle portion was devoted to a brief study of Heegaaard splittings and general-
ized Heegaard splittings. The latter portion touched on a brand new topic: fork
complexes.
During this time Professor Tsuyoshi Kobayashi had raised some interesting ques-
tions about the connectivity properties of generalized Heegaard splittings. The
latter portion of the lecture series was motivated by these questions. And fork
complexes were invented in an eort to illuminate some of the more subtle issues
arising in the study of generalized Heegaard splittings.
In the standard schematic diagram for generalized Heegaard splittings, Heegaard
splittings are stacked on top of each other in a linear fashion. See Figure 1. This
can cause confusion in those cases in which generalized Heegaaard splittings possess
interesting connectivity properties. In these cases, some of the topological features
of the 3-manifold are captured by the connectivity properties of the generalized
Heegaard splitting rather than by the Heegaard splittings of submanifolds into
which the generalized Heegaard splitting decomposes the 3-manifold. See Figure
2. Fork complexes provide a means of description in this context.
Figure 1. The standard schematic diagram
1
2 TOSHIO SAITO, MARTIN SCHARLEMANN AND JENNIFER SCHULTENS
Figure 2. A more informative schematic diagram for a generalized
Heegaard splitting for a manifold homeomorphic to (a surface) S
1
The authors would like to express their appreciation of the hospitality extended
to them during their stay at RIMS. They would also like to thank the many
people that made their stay at RIMS delightful, illuminating and productive, most
notably Professor Hitoshi Murakami, Professor Tsuyoshi Kobayashi, Professor Jun
Murakami, Professor Tomotada Ohtsuki, Professor Kyoji Saito, Professor Makoto
Sakuma, Professor Kouki Taniyama and Dr. Yoav Rieck. Finally, they would like
to thank Dr. Ryosuke Yamamoto for drawing the ne pictures in these lecture
notes.
2. Preliminaries
2.1. PL 3-manifolds. Let M be a PL 3-manifold, i.e., M is a union of 3-simplices
3
i
(i = 1, 2, . . . , t) such that
3
i

3
j
(i = j) is emptyset, a vertex, an edge or a face
and that for each vertex v,
v
3
j
3
j
is a 3-ball (cf. [14]). Then the decomposition
{
3
i
}
1it
of M is called a triangulation of M.
Example 2.1.1. (1) The 3-ball B
3
is the simplest PL 3-manifold in a sense
that B
3
is homeomorphic to a 3-simplex.
(2) The 3-sphere S
3
is a 3-manifold obtained from two 3-balls by attaching their
boundaries. Since S
3
is homeomorphic to the boundary of a 4-simplex, we
see that S
3
is a union of ve 3-simplices. It is easy to show that this gives
a triangulation of S
3
.
Exercise 2.1.2. Show that the following 3-manifolds are PL 3-manifolds.
(1) The solid torus D
2
S
1
.
(2) S
2
S
1
.
(3) The lens spaces. Note that a lens space is obtained from two solid tori by
attaching their boundaries.
Let K be a three dimensional simplicial complex and X a sub-complex of K,
that is, X a union of vertices, edges, faces and 3-simplices of K such that X is a
simplicial complex. Let K
be the second barycentric subdivision of K. A regular

neighborhood of X in K, denoted by (X; K), is a union of the 3-simplices of K
intersecting X (cf. Figure 3).

Proposition 2.1.3. If X is a PL 1-manifold properly embedded in a PL 3-manifold
M (namely, X M = X), then (X; M)
= X B
2
, where X is identied with
LECTURE NOTES ON GENERALIZED HEEGAARD SPLITTINGS 3
X
K
=

(X; K)
Figure 3.
X {a center of B
2
} and (X; M) M is identied with X B
2
(cf. Figure
4).
M
(X; M)
X
Figure 4.
Proposition 2.1.4. Suppose that a PL 3-manifold M is orientable. If X is an
orientable PL 2-manifold properly embedded in M (namely, X M = X), then
(X; M)
= X[0, 1], where X is identied with X{1/2} and (X; M) M is

identied with X [0, 1].
Theorem 2.1.5 (Moise [10]). Every compact 3-manifold is a PL 3-manifold.
In the remainder of these notes, we work in the PL category unless otherwise
specied.
2.2. Fundamental denitions. By the term surface, we will mean a connected
compact 2-manifold.
Let F be a surface. A loop in F is said to be inessential in F if bounds a
disk in F, otherwise is said to be essential in F. An arc properly embedded
in F is said to be inessential in F if cuts o a disk from F, otherwise is said
to be essential in F.
Let M be a compact orientable 3-manifold. A disk D properly embedded in M
is said to be inessential in M if D cuts o a 3-ball from M, otherwise D is said to
be essential in M. A 2-sphere P properly embedded in M is said to be inessential
in M if P bounds a 3-ball in M, otherwise P is said to be essential in M. Let F be
a surface properly embedded in M. We say that F is -parallel in M if F cuts o
a 3-manifold homeomorphic to F [0, 1] from M. We say that F is compressible
in M if there is a disk D M such that DF = D and D is an essential loop
in F. Such a disk D is called a compressing disk. We say that F is incompressible
in M if F is not compressible in M. The surface F is -compressible in M if there
is a disk M such that F is an arc which is essential in F, say , in F
and that M is an arc, say
, with
= . Otherwise F is said to be
-incompressible in M. Suppose that F is homeomorphic neither to a disk nor to
a 2-sphere. The surface F is said to be essential in M if F is incompressible in M
and is not -parallel in M.
Denition 2.2.1. Let M be a connected compact orientable 3-manifold.
(1) M is said to be reducible if there is a 2-sphere in M which does not bound
a 3-ball in M. Such a 2-sphere is called a reducing 2-sphere of M. M is
said to be irreducible if M is not reducible.
(2) M is said to be -reducible if there is a disk properly embedded in M whose
boundary is essential in M. Such a disk is called a -reducing disk.
3. Heegaard splittings
3.1. Denitions and fundamental properties.
Denition 3.1.1. A 3-manifold C is called a compression body if there exists a
closed surface F such that C is obtained from F [0, 1] by attaching 2-handles
along mutually disjoint loops in S {1} and lling in some resulting 2-sphere
boundary components with 3-handles (cf. Figure 5). We denote F {0} by
+
C
and C \
+
C by
C. A compression body C is called a handlebody if
C = .
A compression body C is said to be trivial if C

= F [0, 1].
Denition 3.1.2. For a compression body C, an essential disk in C is called a
meridian disk of C. A union of mutually disjoint meridian disks of C is called
a complete meridian system if the manifold obtained from C by cutting along
are the union of
C [0, 1] and (possibly empty) 3-balls. A complete meridian

system of C is minimal if the number of the components of is minimal among
all complete meridian system of C.
Remark 3.1.3. The following properties are known for compression bodies.
(1) A compression body C is reducible if and only if
C contains a 2-sphere
component.
(2) A minimal complete meridian system of a compression body C cuts C
into
C [0, 1] if
C = , and cuts C into a 3-ball if
C = (hence
C is a handlebody).
(3) By extending the cores of the 2-handles in the denition of the compression
body C vertically to F [0, 1], we obtain a complete meridian system of
C such that the manifold obtained by cutting C along is homeomorphic
to a union of
C [0, 1] and some (possibly empty) 3-balls. This gives a

dual description of compression bodies. That is, a compression body C is
obtained from
C [0, 1] and some (possibly empty) 3-balls by attaching

F [0, 1]
Dual discription

Figure 5.
some 1-handles to
C {1} and the boundary of the 3-balls (cf. Figure

5).
(4) For any compression body C,
C is incompressible in C.
(5) Let C and C
be compression bodies. Suppose that C
is obtained from
C and C
by identifying a component of
C and
+
C
. Then C
is a
compression body.
(6) Let D be a meridian disk of a compression body C. Then there is a
complete meridian system of C such that D is a component of . Any
component obtained by cutting C along D is a compression body.
Exercise 3.1.4. Show Remark 3.1.3.
An annulus A properly embedded in a compression body C is called a spanning
annulus if A is incompressible in C and a component of A is contained in
+
C
and the other is contained in
C.
Lemma 3.1.5. Let C be a non-trivial compression body. Let A be a spanning
annulus in C. Then there is a meridian disk D of C with D A = .
Proof. Since C is non-trivial, there is a meridian disk of C. We choose a meridian
disk D of C such that D intersects A transversely and |D A| is minimal among
all such meridian disks. Note that A
C is an essential loop in the component

of
C containing A
C. We shall prove that D A = . To this end, we

suppose D A = .
Claim 1. There are no loop components of D A.
Proof. Suppose that DA has a loop component which is inessential in A. Let
be a loop component of DA which is innermost in A, that is, cuts o a disk
from A such that the interior of
is disjoint from D. Such a disk
is called an
innermost disk for . We remark that is not necessarily innermost in D. Note
that also bounds a disk in D, say
. Then we obtain a disk D
by applying
cut and paste operation on D with using
and
, i.e., D
is obtained from D
by removing the interior of
and then attaching
(cf. Figure 6). Note that D
is a meridian disk of C. Moreover, we can isotope the interior of D
slightly so
that |D
A| < |DA|, a contradiction. (Such an argument as above is called an

innermost disk argument.)
A
D
=
D
Figure 6.
Hence if DA has a loop component, we may assume that the loop is essential
in A. Let
be a loop component of DA which is innermost in D, and let
be
the innermost disk in D with
. Then
cuts A into two annuli, and let A
be the component obtained by cutting A along
such that A
is adjacent to
C.
Set D
= A
. Then D
( C) is a compressing disk of
C, contradicting (4)
of Remark 3.1.3. Hence we have Claim 1.
Claim 2. There are no arc components of D A.
Proof. Suppose that there is an arc component of DA. Note that D
+
C.
Hence we may assume that each component of D A is an inessential arc in A
whose endpoints are contained in
+
C. Let be an arc component of DA which
is outermost in A, that is, cuts o a disk
from A such that the interior of
is disjoint from D. Such a disk
is called an outermost disk for . Note that

cuts D into two disks

and

(cf. Figure 7).

If both

and

are inessential in C, then D is also inessential in C,

a contradiction. So we may assume that

D =

is essential in C. Then we
can isotope

D slightly so that |
D A| < |D A|, a contradiction. Hence we have

Claim 2. (Such an argument as above is called an outermost disk argument.)
Hence it follows from Claims 1 and 2 that D A = , and this completes the
proof of Lemma 3.1.5.
Remark 3.1.6. Let A be a spanning annulus in a non-trivial compression body
C.
. .
. .
D
Figure 7.
(1) By using the arguments of the proof of Lemma 3.1.5, we can show that
there is a complete meridian system of C with A = .
(2) It follows from (1) above that there is a meridian disk E of C such that
EA = and E cuts o a 3-manifold which is homeomorphic to (a closed
surface) [0, 1] containing A.
Exercise 3.1.7. Show Remark 3.1.6.
Let =
1

p
be a union of mutually disjoint arcs in a compression
body C. We say that is vertical if there is a union of mutually disjoint spanning
annuli A
1
A
p
in C such that
i
A
j
= (i = j) and
i
is an essential arc
properly embedded in A
i
(i = 1, 2, . . . , p).
Lemma 3.1.8. Suppose that =
1

p
is vertical in C. Let D be a
meridian disk of C. Then there is a meridian disk D
of C with D
= which
is obtained by cut-and-paste operation on D. Particularly, if C is irreducible, then
D is ambient isotopic such that D = .
Proof. Let

A = A
1
A
p
be a union of annuli for as above. By using
innermost disk arguments, we see that there is a meridian disk D
such that no
components of D

A are loops which are inessential in

A. We remark that D
is ambient isotopic to D if C is irreducible. Note that each component of

A is
incompressible in C. Hence no components of D

A are loops which are essential
in

A. Hence each component of D

A is an arc; moreover since D is contained
in
+
C, the endpoints of the arc components of D

A are contained in
+
C

A.
Then it is easy to see that there exists an arc
i
( A
i
) such that
i
is essential
in A
i
and
i
D
= . Take an ambient isotopy h

t
(0 t 1) of C such that
h
0
(
i
) =
i
, h
t
(

A) =

A and h
1
(
i
) =
i
(i = 1, 2, . . . , p) (cf. Figure 8). Then the
ambient isotopy h
t
assures that D
is isotoped so that D
is disjoint from .
In the remainder of these notes, let M be a connected compact orientable 3-

manifold.
Denition 3.1.9. Let (
1
M,
2
M) be a partition of -components of M. A
triplet (C
1
, C
2
; S) is called a Heegaard splitting of (M;
1
M,
2
M) if C
1
and C
2
are compression bodies with C
1
C
2
= M,
C
1
=
1
M,
C
2
=
2
M and
C
1
C
2
=
+
C
1
=
+
C
2
= S. The surface S is called a Heegaard surface and the
genus of a Heegaard splitting is dened by the genus of the Heegaard surface.
A
i
i
Figure 8.
K
1
K
2
Figure 9.
Theorem 3.1.10. For any partition (
1
M,
2
M) of the boundary components of
M, there is a Heegaard splitting of (M;
1
M,
2
M).
Proof. It follows from Theorem 2.1.5 that M is triangulated, that is, there is a
nite simplicial complex K which is homeomorphic to M. Let K
be a barycentric
subdivision of K and K
1
the 1-skeleton of K. Here, a 1-skeleton of K is a union
of the vertices and edges of K. Let K
2
K
be the dual 1-skeleton (see Figure 9).

Then each of K
i
(i = 1, 2) is a nite graph in M.
Case 1. M = .
Recall that K
1
consists of 0-simplices and 1-simplices. Set C
1
= (K
1
; M) and
C
2
= (K
2
; M). Note that a regular neighborhood of a 0-simplex corresponds to
a 0-handle and that a regular neighborhood of a 1-simplex corresponds to a 1-
handle. Hence C
1
is a handlebody. Similarly, we see that C
2
is also a handlebody.
Then we see that C
1
C
2
= M and C
1
C
2
= C
1
= C
2
. Hence (C
1
, C
2
; S) is a
Heegaard splitting of M with S = C
1
C
2
.
Case 2. M = .
In this case, we rst take the barycentric subdivision of K and use the same
notation K. Recall that K
is the barycentric subdivision of K. Note that no

3-simplices of K intersect both
1
M and
2
M. Let N(
2
M) be a union of the 3-
simplices in K
intersecting
2
M. Then N(
2
M) is homeomorphic to
2
M[0, 1],
where
2
M {0} is identied with
2
M. Set
2
M =
2
M {1}. Let

K
1
(

K
2
K
1
K
2
2
M
2
M
K
1
K
2
1
M
Figure 10.
resp.) be the maximal sub-complex of K
1
(K
2
resp.) such that

K
1
(

K
2
resp.) is
disjoint from
2
M (
1
M resp.) (cf. Figure 10).
Set C
1
= (
1
M

K
1
; M). Note that C
1
= (
1
M; M) (

K
1
; M). Note again
that a regular neighborhood of a 0-simplex corresponds to a 0-handle and that
a regular neighborhood of a 1-simplex corresponds to a 1-handle. Hence C
1
is
obtained from
1
M[0, 1] by attaching 0-handles and 1-handles and therefore C
1
is a compression body with
C
1
=
1
M. Set C
2
= (N(
2
M)

K
2
; M). By the
same argument, we can see that C
2
C
2
=
2
M. Note
that C
1
C
2
= M and C
1
C
2
= C
1
= C
2
. Hence (C
1
, C
2
; S) is a Heegaard
splitting of M with S = C
1
C
2
.
We now introduce alternative viewpoints to Heegaard splittings as remarks be-
low.
Denition 3.1.11. Let C be a compression body. A nite graph in C is called
a spine of C if C \ (
C )
=
+
C [0, 1) and every vertex of valence one is in
C (cf. Figure 11).

Figure 11.
Remark 3.1.12. Let (C
1
, C
2
; S) be a Heegaard splitting of (M;
1
M,
2
M). Let
i
be a spine of C
i
, and set
i
=
i
M
i
(i = 1, 2). Then
M \ (
2
) = (C
1
\
1
)
S
(C
2
\
2
)
= S (0, 1).
Hence there is a continuous function f : M [0, 1] such that f
1
(0) =
1
,
f
1
(1) =
2
and f
1
(t)
= S (0 < t < 1). This is called a sweep-out picture.

Remark 3.1.13. Let (C
1
, C
2
1
M,
2
M). By a
dual description of C
1
, we see that C
1
is obtained from
1
M[0, 1] and 0-handles
H
0
by attaching 1-handles H
1
. By Denition 3.1.1, C
2
is obtained from S [0, 1]
2
and lling some 2-sphere boundary components with
3-handles H
3
. Hence we obtain the following decomposition of M:
M =
1
M [0, 1] H
0
H
1
S [0, 1] H
2
H
3
.
By collapsing S [0, 1] to S, we have:
M =
1
M [0, 1] H
0
H
1
S
H
2
H
3
.
This is called a handle decomposition of M induced from (C
1
, C
2
; S).
Denition 3.1.14. Let (C
1
, C
2
1
M,
2
M).
(1) The splitting (C
1
, C
2
; S) is said to be reducible if there are meridian disks
D
i
(i = 1, 2) of C
i
with D
1
= D
2
. The splitting (C
1
, C
2
; S) is said to be
irreducible if (C
1
, C
2
; S) is not reducible.
1
, C
2
; S) is said to be weakly reducible if there are meridian
disks D
i
(i = 1, 2) of C
i
with D
1
D
2
= . The splitting (C
1
, C
2
; S) is
said to be strongly irreducible if (C
1
, C
2
; S) is not weakly reducible.
1
, C
2
; S) is said to be -reducible if there is a disk D prop-
erly embedded in M such that DS is an essential loop in S. Such a disk
D is called a -reducing disk for (C
1
, C
2
; S).
1
, C
2
; S) is said to be stabilized if there are meridian disks
D
i
(i = 1, 2) of C
i
such that D
1
and D
2
intersect transversely in a single
point. Such a pair of disks is called a cancelling pair of disks for (C
1
, C
2
; S).
Example 3.1.15. Let (C
1
, C
2
; S) be a Heegaard splitting such that each of
C
i
(i = 1, 2) consists of two 2-spheres and that S is a 2-sphere. Note that there does
not exist an essential disk in C
i
. Hence (C
1
, C
2
; S) is strongly irreducible.
Suppose that (C
1
, C
2
; S) is stabilized, and let D
i
(i = 1, 2) be disks as in (4)
of Denition 3.1.14. Note that since D
1
intersects D
2
transversely in a single
point, we see that each of D
i
(i = 1, 2) is non-separating in S and hence each
of D
i
(i = 1, 2) is non-separating in C
i
. Set C
1
= cl(C
1
\ (D
1
; C
1
)) and C
2
=
C
2
(D
1
; C
1
). Then each of C
i
(i = 1, 2) is a compression body with
+
C
1
=
+
C
2
(cf. (6) of Remark 3.1.3). Set S
=
+
C
1
(=
+
C
2
). Then we obtain the Heegaard
splitting (C
1
, C
2
; S
) of M with genus(S
) = genus(S) 1. Conversely, (C
1
, C
2
; S)
is obtained from (C
1
, C
2
; S
) by adding a trivial handle. We say that (C

1
, C
2
; S) is
obtained from (C
1
, C
2
; S
) by stabilization.
Observation 3.1.16. Every reducible Heegaard splitting is weakly reducible.
Lemma 3.1.17. Let (C
1
, C
2
1
M,
2
M) with
genus(S) 2. If (C
1
, C
2
; S) is stabilized, then (C
1
, C
2
; S) is reducible.
Proof. Suppose that (C
1
, C
2
; S) is stabilized, and let D
i
(i = 1, 2) be meridian
disks of C
i
such that D
1
intersects D
2
transversely in a single point. Then
(D
1
D
2
; S) bounds a disk D
i
in C
i
for each i = 1 and 2. In fact, D
1
(D
2
resp.) is obtained from two parallel copies of D
1
(D
2
resp.) by adding a band
along D
2
\ (the product region between the parallel disks) (D
1
\ (the product
region between the parallel disks) resp.) (cf. Figure 12).
D
1
D
1
D
2
C
1

D
2
D
2
D
1
C
2
Figure 12.
Note that D
1
= D
2
cuts S into a torus with a single hole and the other surface
S
. Since genus(S) 2, we see that genus(S
) 1. Hence D
1
= D
2
is essential
in S and therefore (C
1
, C
2
; S) is reducible.
Denition 3.1.18. Let (C
1
, C
2
1
M,
2
M).
(1) Suppose that M

= S
3
. We call (C
1
, C
2
; S) a trivial splitting if both C
1
and
C
2
are 3-balls.
(2) Suppose that M
= S
3
. We call (C
1
, C
2
; S) a trivial splitting if C
i
is a trivial
handlebody for i = 1 or 2.
Remark 3.1.19. Suppose that M
= S
3
. If (M;
1
M,
2
M) admits a trivial split-
ting (C
1
, C
2
; S), then it is easy to see that M is a compression body. Particularly,
if C
2
(C
1
resp.) is trivial, then
M =
1
M and
+
M =
2
M (
M =
2
M and
+
M =
1
M resp.).
Lemma 3.1.20. Let (C
1
, C
2
; S) be a non-trivial Heegaard splitting of (M;
1
M,
2
M).
If (C
1
, C
2
; S) is -reducible, then (C
1
, C
2
; S) is weakly reducible.
Proof. Let D be a -reducing disk for (C
1
, C
2
; S). (Hence DS is an essential loop
in S.) Set D
1
= DC
1
and A
2
= DC
2
. By exchanging subscripts, if necessary,
we may suppose that D
1
is a meridian disk of C
1
and A
2
is a spanning annulus in
C
2
. Note that A
2
C
2
is an essential loop in the component of
C
2
containing
A
2

C
2
. Since C
2
is non-trivial, there is a meridian disk of C
2
. It follows from
Lemma 3.1.5 that we can choose a meridian disk D
2
of C
2
with D
2
A
2
= . This
implies that D
1
D
2
= . Hence (C
1
, C
2
; S) is weakly reducible.
3.2. Hakens theorem. In this subsection, we prove the following.
Theorem 3.2.1. Let (C
1
, C
2
1
M,
2
M).
(1) If M is reducible, then (C
1
, C
2
; S) is reducible or C
i
is reducible for i = 1
or 2.
(2) If M is -reducible, then (C
1
, C
2
; S) is -reducible.
Note that the statement (1) of Theorem 3.2.1 is called Hakens theorem and
proved by Haken [4], and the statement (2) of Theorem 3.2.1 is proved by Casson
and Gordon [1].
We rst prove the following proposition, whose statement is weaker than that
of Theorem 3.2.1, after showing some lemmas.
Proposition 3.2.2. If M is reducible or -reducible, then (C
1
, C
2
; S) is reducible,
-reducible, or C
i
is reducible for i = 1 or 2.
We give a proof of Proposition 3.2.2 by using Otals idea (cf. [11]) of viewing
the Heegaard splittings as a graph in the three dimensional space.
Edge slides of graphs. Let be a nite graph in a 3-manifold M. Choose an
edge of . Let p
1
and p
2
be the vertices of incident to . Set

= \ . Here,
we may suppose that (
; M) consists of two points, say p

1
and p
2
, and that
cl( \ (p
1
p
2
)) consists of
0
,
1
and
2
with
0
= p
1
p
2
,
1
= p
1
p
1
and
2
= p
2
p
2
(cf. Figure 13).
p
1
p
2
p
1
p
2
Figure 13.
Take a path on (
; M) with p
1
. Let be an arc obtained from

0
2
by adding a straight short arc in (
; M) connecting the endpoint of

other than p
1
and a point p
1
in the interior of an edge of

(cf. Figure 14). Let
be a graph obtained from

by adding p
1
as a vertex. Then we say that
is obtained from by an edge slide on .
0
..
p
1
p
2
p
1
Figure 14.
If p
1
is a trivalent vertex, then it is natural for us not to regard p
1
as a vertex of
. Particularly, the deformation of which is depicted as in Figure 15 is realized

by an edge slide and an isotopy. This deformation is called a Whitehead move.
A Proof of Proposition 3.2.2. Let be a spine of C
1
. Note that (
C
1
; M)
is obtained from regular neighborhoods of
C
1
and the vertices of by attaching
Figure 15.
1-handles corresponding to the edges of . Set
= (; M). The notation h

0
v
,
called a vertex of
, means a regular neighborhood of a vertex v of . Also, the

notation h
1
, called an edge of
, means a 1-handle corresponding to an edge of

. Let = D
1
D
k
be a minimal complete meridian system of C
2
.
Let P be a reducing 2-sphere or a -reducing disk of M. If P is a -reducing
disk, we may assume that P
C
2
by changing subscripts. We may assume
that P intersects and transversely. Set = P (
). We note that is a
union of disks P
and a union of arcs and loops P in P. We choose P,

and so that the pair (|P |, |P |) is minimal with respect to lexicographic
order.
Lemma 3.2.3. Each component of P is an arc.
Proof. For some disk component, say D
1
, of , suppose that P D
1
has a loop
component. Let be a loop component of P D
1
which is innermost in D
1
,
and let
be an innermost disk for . Let
be a disk in P with
= . Set
P
= (P \
if P is a -reducing disk, or set P
= (P \
and P
if P is a reducing 2-sphere. If P is a -reducing disk, then P
is also a -reducing
disk. If P is a reducing 2-sphere, then either P
or P
, say P
, is a reducing 2-
sphere. Moreover, we can isotope P
so that (|P
|, |P
|) < (|P |, |P |).

This contradicts the minimality of (|P |, |P |).
By Lemma 3.2.3, we can regard as a graph in P which consists of fat-vertices
P
and edges P . An edge of the graph is called a loop if the edge joins
a fat-vertex of to itself, and a loop is said to be inessential if the loop cuts o a
disk from cl(P \
) whose interior is disjoint from
.
Lemma 3.2.4. does not contain an inessential loop.
Proof. Suppose that contains an inessential loop . Then cuts o a disk
from cl(P \
) such that the interior of
is disjoint from
(cf. Figure 16).

We may assume that
D
1
. Then cuts D
1
into two disks D
1
and
D
1
(cf. Figure 17).
Let C
2
be the component, which is obtained by cutting C
2
along , such that
C
2
contains
. Let D
+
1
be the copy of D
1
in C
2
with D
+
1

= and D
1
the
other copy of D
1
. Note that C
2
is a 3-ball or a (a component of
C
2
)[0, 1]. This
an inessential monogon
Figure 16.
P P
=
D
1
Figure 17.
shows that there is a disk
in
+
C
2
such that
and
bounds
a 3-ball in C
2
. Note that
D
+
1
= . By changing superscripts, if necessary, we
may assume that
1
(cf. Figure 18).
+
C
2
D
+
1
..
D
1
D
1
. .
Figure 18.
Set D
0
=
1
if
1
= , and D
0
=
1
if
1
= . We
may regard D
0
as a disk properly embedded in C
2
. Set
= D
0
D
2
D
k
.
Then we see that
is a minimal complete meridian system of C

2
. We can further
isotope D
0
slightly so that |P
| < |P |. This contradicts the minimality of

(|P |, |P |).
A fat-vertex of is said to be isolated if there are no edges of adjacent to the
fat-vertex (cf. Figure 19).
an isolated fat-vertex
Figure 19.
Lemma 3.2.5. If has an isolated fat-vertex, then (C
1
, C
2
; S) is reducible or
-reducible.
Proof. Suppose that there is an isolated fat-vertex D
v
of . Recall that D
v
is a
component of P
which is a meridian disk of C

1
. Note that D
v
is disjoint from
(cf. Figure 20).
P
D
v
Figure 20.
Let C
2
be the component obtained by cutting C
2
along such that C
2
contains
D
v
. If D
v
bounds a disk D
v
in C
2
, then D
v
and D
v
indicates the reducibility of
(C
1
, C
2
; S). Otherwise, C
2
is a (a closed orientable surface) [0, 1], and D
v
is a
boundary component of a spanning annulus in C
2
(and hence C
2
). Hence we see
that (C
1
, C
2
; S) is -reducible.
Lemma 3.2.6. Suppose that no fat-vertices of are isolated. Then each fat-vertex
of is a base of a loop.
Proof. Suppose that there is a fat-vertex D
w
of which is not a base of a loop.
Since no fat-vartices of are isolated, there is an edge of adjacent to D
w
. Let
be the edge of with h
1
D
w
. (Recall that h
1
is a 1-handle of
corresponding
to .) Let D be a component of with D h
1
= . Let C
w
be a union of the
arc components of D P which are adjacent to D
w
. Let be an arc component
of C
w
which is outermost among the components of C
w
. We call such an arc an
outermost edge for D
w
of . Let
D be a disk obtained by cutting D along

whose interior is disjoint from the edges incident to D
w
. We call such a disk
an outermost disk for (D

w
, ). (Note that
may intersects P transversely (cf.

Figure 21).) Let D
w
(= D
w
) be the fat-vertex of attached to . Then we have
the following three cases.
D
1
D
w
D
w
D
w
D
w
D
w
D
w
D
w
D
w
Figure 21.
Case 1. (
\ ) (h
1
D).
In this case, we can isotope along
to reduce |P | (cf. Figure 22).
D
w
D
w
D
w
D
w
=
Figure 22.
Case 2. (
\ ) (h
1
D) and D
w
(h
1
D).
Let p be the vertex of such that p = and h
0
p

= . Let be the
component of cl( \ D
w
) which satises p = . Then we can slide along
so that contains (cf. Figure 23). We can further isotope slightly to reduce
|P |, a contradiction.
Case 3. (
\ ) (h
1
D) and D
w
(h
1
D).
Let p and p
be the endpoints of . Let and
be the components of cl( \

(D
w
D
w
)) which satisfy p = and p
= . Suppose rst that p = p
.
Then we can slide along
so that contains (cf. Figure 24). We can further

isotope slightly to reduce |P |, a contradiction.
Suppose next that p = p
. In this case, we perform the following operation which

is called a broken edge slide.
D
w
P
p
D
w
D
w
=
Figure 23.
D
w
D
w
p
p
D
w
D
w
p p
=
Figure 24.
P
D
w
D
w
p
w
=
Figure 25.
We rst add w
= D
w
as a vertex of . Then w
cuts into two edges
and cl( \
). Since is an outermost edge for D

w
of , we see that
(cf.
Figure 25). Hence we can slide cl( \
) along
so that cl( \
) contains . We
now remove the verterx w
of , that is, we regard a union of
and cl( \
) as
an edge of again. Then we can isotope cl( \
) slightly to reduce |P |, a
contradiction (cf. Figure 26).
Proof of Proposition 3.2.2. By Lemma 3.2.5, if there is an isolated fat-vertex of ,

then we have the conclusion of Proposition 3.2.2. Hence we suppose that no fat-
vertices of are isolated. Then it follows from Lemma 3.2.6 that each fat-vertex
p
w
=
Figure 26.
of is a base of a loop. Let be a loop which is innermost in P. Then cuts
a disk
from cl(P \
). Since is essential (cf. Lemma 3.2.4), we see that
contains a fat-vertex of . But since is innermost, such a fat-vertex is not a base

of any loop. Hence such a fat-vertex is isolated, a contradiction. This completes
the proof of Proposition 3.2.2.
Proof of (1) in Theorem 3.2.1. Suppose that M is reducible. Then by Proposition
3.2.2, we see that (C
1
, C
2
; S) is reducible or -reducible, or C
i
is reducible for i = 1
or 2. If (C
1
, C
2
; S) is reducible or C
i
is reducible for i = 1 or 2, then we are done. So
we may assume that C
1
and C
2
are irreducible and that (C
1
, C
2
; S) is -reducible.
By induction on the genus of the Heegaard surface S, we prove that (C
1
, C
2
; S) is
reducible.
Suppose that genus(S) = 0. Since C
i
(i = 1, 2) are irreducible, we see that each
of C
i
(i = 1, 2) is a 3-ball. Hence M is the 3-sphere and therefore M is irreducible,
a contradiction. So we may assume that genus(S) > 0. Let P be a -reducing
disk of M with |P S| = 1. By changing subs cripts, if necessary, we may assume
that P C
1
= D is a disk and P C
2
= A is a spanning annulus.
Suppose that genus(S) = 1. Since C
i
(i = 1, 2) are irreducible, we see that
C
i
contain no 2-sphere components. Since C
1
contains an essential disk D, we
see that C
1

= D
2
S
1
. Since C
2
contains a spanning annulus A, we see that
C
2

= T
2
[0, 1]. It follows that M

= D
2
S
1
and hence M is irreducible, a
contradiction.
Suppose that genus(S) > 1. Let C
1
(C
2
resp.) be the manifold obtained from
C
1
(C
2
resp.) by cutting along D (A resp.), and let A
+
and A
be copies of A in
C
2
. Then we see that C
1
consists of either a compression body or a union of two
compression bodies (cf. (6) of Remark 3.1.3). Let C
2
be the manifold obtained
from C
2
by attaching 2-handles along A
+
and A
. It follows from Remark 3.1.6

that C
2
consists of either a compression body or a union of two compression bodies.
Suppose that C
1
consists of a compression body. This implies that C
2
consists
of a compression body (cf. Figure 27). We can naturally obtain a homeomorphism
+
C
1

+
C
2
from the homeomorphism
+
C
1

+
C
2
. Set
+
C
1
=
+
C
2
= S
.
Then (C
1
, C
2
; S
) is a Heegaard splitting of the 3-manifolds M
obtained by cutting
M along P. Note that genus(S
) = genus(S) 1. Moreover, by using innermost

disk arguments, we see that M
is also reducible.
Claim.
D
S
C
1
C
2
A
1
C
2
Figure 27.
(1) If C
1
is reducible, then C
1
is reducible.
(2) If C
2
is reducible, then one of the following holds.
(a) C
2
is reducible.
(b) The component of
C
2
intersecting A is a torus, say T.
Proof. Exercise 3.2.7.
Recall that we assume that C
i
(i = 1, 2) are irreducible. Hence it follows from
(1) of the claim that C
1
is irreducible. Also it follows from (2) of the claim that
either (I) C
2
is irreducible or (II) C
2
is reducible and the condition (b) of (2) in
Claim 1 holds.
Suppose that the condition (I) holds. Then by induction on the genus of a
Heegaard surface, (C
1
, C
2
; S
) is reducible, i.e., there are meridian disks D

1
and
D
2
of C
1
and C
2
respectively with D
1
= D
2
. Note that this implies that C
i
(i = 1, 2) are non-trivial. Let
+
and
be the co-cores of the 2-handles attached

to C
2
. Then we see that
+

is vertical in C
2
. It follows from Lemma 3.1.8
that we may assume that D
2
(
+
) = , i.e., D
2
is disjoint from the 2-handles.
Hence the pair of disks D
1
and D
2
survives when we restore C
1
and C
2
from C
1
and C
2
respectively. This implies that (C
1
, C
2
; S) is reducible and hence we obtain
the conclusion (1) of Theorem 3.2.1.
Suppose that the condition (II) holds. Then it follows from (2) of Remark 3.1.6
that there is a separating disk E
2
in C
2
such that E
2
is disjoint from A and that
E
2
cuts o T
2
[0, 1] from C
2
with T
2
{0} = T. Let be a loop in S (T
2
{1})
which intersects AS(= AS = D) in a single point. Let E
1
be a disk properly
embedded in C
1
which is obtained from two parallel copies of D by adding a band
along \ (the product region between the parallel disks). Since genus(S) > 1, we
see that E
1
is a separating meridian disk of C
1
. Since E
1
is isotopic to E
2
, we
see that (C
1
, C
2
; S) is reducible. Hence we obtain the conclusion (1) of Theorem
3.2.1.
The case that C
1
is a union of two compression bodies is treated analogously,
and we leave the proof for this case to the reader (Exercise 3.2.8).
Exercise 3.2.7. Show the claim in the proof of Theorem 3.2.1.
Exercise 3.2.8. Prove that the conclusion (1) of Theorem 3.2.1 holds in case that
C
1
consists of two compression bodies.
Proof of (2) in Theorem 3.2.1. Suppose that M is -reducible. If (C
1
, C
2
; S) is -
reducible, then we are done. Let

C
i
be the compression body obtained by attaching
3-balls to the 2-sphere boundary components of C
i
(i = 1, 2). Set

M =

C
1

C
2
. Then

M is also -reducible. Then it follows from (1) of Remark 3.1.3 and
Proposition 3.2.2 that (
C
1
,

C
2
; S) is reducible or -reducible. If (
C
1
,

C
2
; S) is -
reducible, then we see that (C
1
, C
2
; S) is also -reducible. Hence we may assume
that (
C
1
,

C
2
; S) is reducible. By induction on the genus of a Heegaard surface,
we prove that (C
1
, C
2
; S) is -reducible. Let P
be a reducing 2-sphere of

M with
|P
S| = 1. For each i = 1 and 2, set D

i
= P

C
i
, and let

C
i
be the manifold
obtained by cutting

C
i
along D
i
, and let D
+
i
and D
i
be copies of D
i
in

C
i
. Then
each of

C
i
(i = 1, 2) is either (1) a compression body if D
i
is non-separating in

C
i
or (2) a union of two compression bodies if D
i
is separating in

C
i
. Note that we
can naturally obtain a homeomorphism
+

C
1

+

C
2
from the homeomorphism
+

C
1

+

C
2
. Set

M
=

C
1

C
2
and
+

C
1
=
+

C
2
= S
. Then (
1
,

C
2
; S
) is
either (1) a Heegaard splitting or (2) a union of two Heegaard splittings (cf. Figure
28).
D
1
C
1

C
2
D
2
2
Figure 28.
By innermost disk arguments, we see that there is a -reducing disk of

M disjoint
from P
. This implies that a component of

M
is -reducible and hence one of the

Heegaard splittings of (
1
,

C
2
; S
) is -reducible. By induction on the genus of a

Heegaard surface, we see that (
C
1
,

C
2
; S) is -reducible. Therefore (C
1
, C
2
; S) is
also -reducible and hence we have (2) of Theorem 3.2.1.
3.3. Waldhausens theorem. We devote this subsection to a simplied proof of
the following theorem originally due to Waldhausen [21]. To prove the theorem,
we exploit Gabais idea of thin position (cf. [3]), Johannsons technique (cf. [6])
and Otals idea (cf. [11]) of viewing the Heegaard splittings as a graph in the three
dimensional space.
Theorem 3.3.1 (Waldhausen). Any Heegaard splitting of S
3
is standard, i.e., is
obtained from the trivial Heegaard splitting by stabilization.
Thin position of graphs in the 3-sphere. Let S
3
be a nite graph in
which all vertices are of valence three. Let h : S
3
[1, 1] be a height function
such that h
1
(t) = P(t)

= S
2
for t (1, 1), h
1
(1) = (the south pole of S
3
),
and h
1
(1) = (the north pole of S
3
). Let V denote the set of vertices of .
Denition 3.3.2. The graph is in Morse position with respect to h if the
following conditions are satised.
(1) h|
\V
has nitely many non-degenerate critical points.
(2) The height of critical points of h|
\V
and the vertices V are mutually dif-
ferent.
A set of the critical heights for is the set of height at which there is either a
critical point of h|
\V
or a component of V. We can deform by an isotopy so that
a regular neighborhood of each vertex v of is either of Type-y (i.e., two edges
incident to v is above v and the remaining edge is below v) or of Type- (i.e., two
edges incident to v is below v and the remaining edge is above v). Such a graph
is said to be in normal form. We call a vertex v a y-vertex (a -vertex resp.) if
(v; ) is of Type-y (Type- resp.).
Suppose that is in Morse position and in normal form. Note that (; S
3
) can
be regarded as the union of 0-handles corresponding to the regular neighborhood
of the vertices and 1-handles corresponding to the regular neighborhood of the
edges. A simple loop in (; S
3
) is in normal form if the following conditions
are satised.
(a) For each 1-handle (
= D
2
[0, 1]), each component of (D
2
[0, 1]) is an
essential arc in the annulus D
2
[0, 1].
(b) For each 0-handle (
= B
3
), each component of B
3
is an arc which is
essential in the 2-sphere with three holes cl(B
3
\ (the 1-handles incident to B
3
)).
Let D be a disk properly embedded in cl(S
3
\ (; S
3
)). We say that D is in
normal form if the following conditions are satised.
(1) D is in normal form.
(2) Each critical point of h|
int(D)
is non-degenerate.
(3) No critical points of h|
int(D)
occur at critical heights of .
(4) No two critical points of h|
int(D)
occur at the same height.
(5) h|
D
is a Morse function on D satisfying the following (cf. Figure 29).
(a) Each minimum of h|
D
occurs either at a y-vertex in half-center
singularity or at a minimum of in half-center singularity.
(b) Each maximum of h|
D
occurs either at a -vertex in half-center
singularity or at a maximum of in half-center singularity.
By Morse theory (cf. [9]), it is known that D can be put in normal form.
Recall that h : S
3
[1, 1] is a height function such that h
1
(t) = P(t)

= S
2
for t (1, 1), h
1
(1) = (the south pole of S
3
), and h
1
(1) = (the north pole of
S
3
). We isotope to be in Morse position and in normal form. For t (1, 1), set
w
(t) = |P(t)|. Note that w
(t) is constant on each component of (1, 1)\(the

critical heights of ). Set W
= max{w
(t)|t (1, 1)} (cf. Figure 30).

Let n
be the number of the components of (1, 1) \ (the critical heights of )

on which the value W
is attained.
Denition 3.3.3. A graph S
3
is said to be in thin position if (W
, n
) is
minimal with respect to lexicographic order among all graphs which are obtained
Figure 29.
c
6
c
5
c
4
c
3
c
2
c
1
c
0
P(t)
Figure 30.
from by ambient isotopies and edge slides and are in Morse position and in
normal form.
A proof of Theorem 3.3.1. Let (C
1
, C
2
; S) be a genus g > 0 Heegaard splitting
of S
3
. Let be a trivalent spine of C
1
. Note that (
C
1
; M) is obtained
from regular neighborhoods of
C
1
and the vertices of by attaching 1-handles
corresponding to the edges of . Set
= (; M). As in Section 3.2, the notation

h
0
v
, called a vertex of
, means a regular neighborhood of a vertex v of . Also,

the notation h
1
, called an edge of
, means a 1-handle corresponding to an edge

of . Let
1
(
2
resp.) be a complete meridian system of C
1
(C
2
resp.).
Proposition 3.3.4. There is an edge of
which is disjoint from

2
, or is
modied by edge slides so that the modied graph contains an unknotted cycle
(i.e., the modied graph contains a graph so that bounds a disk in S
3
).
Proof of Theorem 3.3.1 via Proposition 3.3.4. We prove Theorem 3.3.1 by induc-
tion on the genus of a Heegaard surface. If genus(S) = 0, then (C
1
, C
2
; S) is
standard (cf. Denition 3.1.18). So we may assume that genus(S) > 0 for a
Heegaard splitting (C
1
, C
2
; S).
Suppose rst that has an unknotted cycle . Then (; C
1
) is a standard
solid torus in S
3
, that is, the exterior of (; C
1
) is a solid torus. Since C
1
=
cl(C
1
\ (; C
1
)) is a compression body, we see that (C
1
, C
2
; S) is a Heegaard
splitting of the solid torus cl(S
3
\ (; C
1
)). Since a solid torus is -reducible,
(C
1
, C
2
; S) is -reducible by Theorem 3.2.1, that is, there is a -reducing disk
D
for (C
1
, C
2
; S) with |D
S| = 1. Since (; C
1
) is a standard solid torus in
S
3
, D
intersects a meridian disk D
of (; C
1
) transversely in a single point.
Set D
2
= D
C
2
. Then by extending D
, we obtain a meridian disk D

1
of C
1
such that D
1
intersects D
2
transversely in a single point, i.e., D
1
and D
2
give
stabilization of (C
1
, C
2
; S). Hence we obtain a Heegaard splitting (C
1
, C
2
; S
) with
genus(S
) < genus(S) (cf. Figure 31). By induction on the genus of a Heegaard

surface, we can see that (C
1
, C
2
; S) is standard.
C
D
1
D
2
Figure 31.
Suppose next that there is an edge of with h
1

2
= . Let D
be a
meridian disk of C
1
which is co-core of the 1-handle h
1
. Note that D

2
= .
Cutting C
2
along
2
, we obtain a union of 3-balls and hence we see that D
bounds a disk, say D
, properly embedded in one of the 3-balls. Note that D
corresponds to a meridian disk of C

2
. Hence we see that (C
1
, C
2
; S) is reducible. It
follows from a generalized Schonies theorem that every 2-sphere in S
3
separates it
into two 3-balls (cf. Section 2.F.5 of [13]). Hence by cutting S
3
along the reducing
2-sphere and capping o 3-balls, we obtain two Heegaard splittings of S
3
such
that the genus of each Heegaard surface is less than that of S. Then we see that
(C
1
, C
2
; S) is standard by induction on the genus of a Heegaard surface.
In the remainder, we prove Proposition 3.3.4. Let h : S
3
[1, 1] be a height
function such that h
1
(t) = P(t)

= S
2
for t (1, 1), h
1
(1) = (the south
pole of S
3
), and h
1
(1) = (the north pole of S
3
). We may assume that is in
thin position. We also assume that each component of
2
is in normal form,
1
intersects
2
transversely and |
1

2
| is minimal.
For the proof of Proposition 3.3.4, it is enough to show the following; if there are
no edges of
which are disjoint from

2
, then is modied by edge slides so that
the modied graph contains an unknotted cycle. Hence we suppose that there are
no edges of

2
. Set (t) = P(t) (

2
). We note
that P(t), and
2
intersect transversely at a regular height t. In the following,
we mainly consider such a regular height t with (t) = unless otherwise denoted.
We also note that we may assume that (t) does not contain a loop component
by an argument similar to the proof of Lemma 3.2.3. Hence (t) is regarded as a
graph in P(t) which consists of fat-vertices P(t)
and edges P(t)

2
.
Lemma 3.3.5. If there is a fat-vertex of (t) with valence less than two, then
is modied by edge slides so that the modied graph contains an unknotted cycle.
v
of (t) with valence less than two.
Let be the edge of with h
1
D
v
and p one of the endpoints of . Since
we assume that there are no edges of

2
, we see that
any fat-vertex of (t) is of valence greater than zero. Hence D
v
is of valence one.
Then there is the disk component D of
2
with h
1
D = . Since D intersects
the fat-vertex D
v
in a single point and hence D intersects h
1
in a single arc,
we can perform an edge slide on along cl(D \ h
1
) to obtain a new graph
from (cf. Figure 32). Clearly,
contains an unknotted cycle (bounding a disk

corresponding to D
2
).
P(t)
D
Figure 32.
An edge of a graph (t) is said to be simple if the edge joins distinct two fat-
vertices of (t). Recall that an edge of a graph (t) is called a loop if the edge is
not simple.
Lemma 3.3.6. Suppose that there are no fat-vertices of valence less than two.
Then there exists a fat-vertex D
w
of (t) such that any outermost edge for D
w
of
(t) is simple.
Proof. If (t) does not contain a loop, then we are done. So we may assume that
(t) contains a loop, say . Let D
v
be a fat-vertex of (t) which is a bese of
. By an argument similar to the proof of Lemma 3.2.4, we can see that cuts
cl(P(t) \ D
v
) into two disks, and each of the two disks contains a fat-vertex of
(t). Let
0
be a loop of (t) which is innermost in P(t). Let D
w
be a fat-vertex
contained in the interior of the innermost disk bounded by
0
. Note that D
w
is
not isolated and that every edge contained in the interior of the innermost disk is
simple. Hence any outermost edge for D
w
of (t) is simple.
Let D
w
be a fat-vertex of (t) with a simple edge ( (t)). We may assume
that is a simple outermost edge for D
w
of (t) and is contained in a disk
component D of
2
. It follows from Lemma 3.3.6 that we can always nd such a
fat-vertex D
w
and an edge if each fat-vertex of (t) is of valence greater than
one. Let
be the outermost disk for (D

w
, ). We say an outermost edge is
upper (lower resp.) if (;
) is above (below resp.) with respect to the height

function h. Let t
0
be a regular height with w
(t
0
) = W
.
Lemma 3.3.7. Let D
w
be a fat-vertex of (t
0
) with a simple outermost edge for
D
w
of (t). Then we have one of the following.
(1) All the simple outermost edges for D
w
of (t) are either upper or lower.
(2) is modied by edge slides so that the modied graph contains an unknotted
cycle.
Proof. Suppose that (t) contains simple outermost edges for D
w
, say and
,
such that is upper and
is lower. For the proof of Lemma 3.3.7, it is enough to

show that is modied by edge slides so that there is an unknotted cycle. Let
and

w
, ) and (D
w
,
) respectively. Let be the

edge of with h
1
D
w
. Let (
resp.) be a union of the components obtained

by cutting by the two fat-vertices of (t
0
) incident to (
resp.) such that a

1-handle correponding to each component intersects
\ (
resp.). We
note that (
resp.) satises one of the following conditions.

(1) (
resp.) consists of an arc such that (
resp.) and share a single

endpoint.
(2) (
resp.) consists of an arc with int() (
int() resp.).
(3) (
resp.) consists of two subarcs of such that each component of (
resp.) and share a single endpoint.

In each of the conditions above, corresponding gures are illustrated in Figure
33. (We remark that there is a case in the condition (3) which is similar to the
latter half of Case 3 in the proof of Lemma 3.2.6. Recall that we only have to use
broken edge slides in that case. Hence we omit details in such a case here.)
Case (1)-(1). Both and
satisfy the condition (1).

If the endpoints of are the same as those of
, then we can slide (
resp.) to
(
resp.) along the disk
resp.) and hence we obtain an unknotted cycle

(cf. Figure 34). Otherwise, we can perform a Whitehead move on to reduce
(W
, n
), a contradiction (cf. Figure 35).

Case (1)-(2). Either or
, say , satises the condition (1) and
satises the
condition (2).
Then we can slide (
resp.) to (
resp.). Then
is further isotoped to reduce (W
, n

I
II
III

D
w
D
w
D
w
Figure 33.
P(t
0
)
=
Figure 34.
satises the
condition (3).
Let
1
and
2
be the components of
with h
1

1
D
w
. Note that
2
and
hence int( )
2
. This implies that int( ) P(t
0
) = . Hence we can slide
= =
Figure 35.
=

D
w

..

..
w
Figure 36.
to along the disk
. We can further isotope slightly to reduce (W
, n
), a
=

2
D
w

..
.. . .
w

1
Figure 37.

Then we can slide (
resp.) to (
resp.).
Moreover, we can isotope slightly to reduce (W
, n
), a contradiction (cf. Figure

59).
satises the
condition (3).
=
P(t
0
)
D
w
w
..

..

Figure 38.
Note that
consists of two arcs, say
1
and
2
, with
1
= D
w
. Then we
have the following cases.
(i)
2
is disjoint from . In this case, we can slide (
resp.) to (
resp.) along
the disk
resp.). Moreover, we can isotope slightly to reduce (W
, n
), a
=
1
D
w
w..

.. ..

2
Figure 39.
(ii)
2
consists of a point, i.e.,
2
and share one endpoint. Note that

= and
= . In this case, we can slide (
resp.) to (
resp.)
along the disk
resp.) and hence we obtain an unknotted cycle (cf. Figure

62).
=
1
D
w
w

..
.. ..

2
Figure 40.
(iii)
2
consists of an arc. In this case, we can slide to along the disk
.
Since
2
consists of an arc, contains at least three critical points. Hence we
can further isotope slightly to reduce (W
, n

=
D
w
w ..

.. . .

2
Figure 41.

Let
1
and
2
(
1
and
2
resp.) be the components of (
resp.) with h
1

1
D
w
(h
1

1
D
w
resp.). Note that
1

2
and
1

2
. In this case, we can slide
1
to
along the disk
. Since
1

2
,
1
contains at least one critical point. Hence
we can further isotope slightly to reduce (W
, n

42).
=
D
w
w

1

2
.. ..
.. . .

2
Figure 42.
Suppose that D
w
is a fat-vertex of (t
0
) such that there are no loops based on
D
w
. It follows from Lemma 3.3.7 that all the simple outermost edges for D
w
of
(t
0
) are either upper or lower.
Lemma 3.3.8. Suppose that all of the simple outermost edges for D
w
of (t
0
) are
upper (lower resp.). Then one of the following holds.
(1) For each fat-vertex D
w
of (t
0
), every simple outermost edges for D
w
of
(t
0
) is upper (lower resp.).
(2) is modied by edge slides so that the modied graph contains an unknotted
cycle.
Proof. Since the arguments are symmetric, we may suppose that all the simple
outermost edges for D
w
of (t
0
) are upper. Let be a simple outermost edge for
D
w
of (t
0
). Note that is upper. Suppose that there is a fat-vertex D
w
such
that (t
0
) contains a lower simple outermost edge
for D
w
. Let
resp.) be
the outermost disk for (D
w
, ) ((D
w
,
) resp.). Let (
resp.) be the edge of

with h
1
D
w
(h
1
D
w
resp.). Let (
resp.) be a union of the components

obtained by cutting by the two fat-vertices of (t
0
) incident to (
resp.) such
that a 1-handle correponding to each component intersects
\ (
resp.).
Then (
resp.) satises one of the conditions (1), (2) and (3) in the proof of
Lemma 3.3.7. The proof of Lemma 3.3.8 is divided into the following cases.
Case A.
= .
Then we have the following six cases. In each case, we can slide (a component
of) (
resp.) to (
resp.). Moreover, we can

isotope and
slightly to reduce (W
, n
) is reduced, a contradiction.
Case A-(1)-(1). Both and

See Figure 43.
=
Figure 43.
Case A-(1)-(2). Either or
satises
the condition (2).
See Figure 44.
=
Figure 44.
satises
the condition (3).
=
Figure 45.
See Figure 45.

See Figure 46.
=
Figure 46.
satises
the condition (3).
See Figure 47.
=
Figure 47.

See Figure 48.
=
Figure 48.
Case B.
= .
Case B-(1)-(1). Both and

We rst suppose that int( )int(
) = . Then we can slide (
resp.) to (
resp.). If =
(= {w, w
}), then
composes
an unknotted cycle and hence Lemma 3.3.8 holds (cf. Figure 49). Otherwise,
we can perform a Whitehead move on and hence we can reduce (W
, n
), a
P(t
0
)
=
w

..

..
Figure 49.
We next suppose that int( ) int(
) = . Then there are two possibilities: (1)

or
, say the latter holds and (2)
and
. In each case, we
can slide to along the disk
. Moreover, we can isotope slightly to reduce

(W
, n
), a contradiction (cf. Figures 51 and 52).

Case B-(1)-(2). Either or
satises
the condition (2).
We rst suppose that int( ) int(
resp.) to
(
resp.). Moreover, we can isotope slightly to

reduce (W
, n

=
D
w
w

..

..
Figure 50.
=

w

..
w
. .

Figure 51.
We next suppose that int( )int(
) = . Note that it is impossible that
.
Hence there are two possibilities:
and
. In each case, we can slide

to along the disk
. Moreover, we can isotope slightly to reduce (W
, n
), a
contradiction (cf. Figures 54 and 55).
satises
the condition (3).
Let
1
and
2
with h
1

1
D
w
.
We rst suppose that
1
. Then we can slide
1
into
along the disk
.
Moreover, we can isotope slightly to reduce (W
, n

56).
=

..
w
. .

Figure 52.
=

D
w

..

..
w
Figure 53.
=
w

..
..

Figure 54.
We next suppose that
1
. Then there are two possibilities:
2
= and

2
= . In each case, we can slide to along the disk
. Moreover, we can
isotope slightly to reduce (W
, n

=
w

..
..

Figure 55.
=
w

..
. .

1
..

2
Figure 56.

resp.) to
(
resp.). Moreover, we can isotope slightly to

reduce (W
, n

We next suppose that int( ) int(
) = . Then there are two possibilities: (1)

or
, say the latter holds and (2)
and
. In each case, we
. Moreover, we can isotope slightly to reduce

(W
, n

satises
the condition (3).
Let
1
and
2
with
1
D
w
.
) = . Since

= , we may suppose
that
1
(=
1
) consists of a single point. Then we can slide (
1
resp.)
to (
resp.). If
2
= , then
2
=
2

consists of a single point. Hence
1
composes an unknotted cycle and hence
=
1
.. ..

2
. .

Figure 57.
=

..
..

1
. .

2
Figure 58.
=
P(t
0
)
D
w
w
..

..

Figure 59.
Lemma 3.3.8 holds (cf. Figure 62). Otherwise, we can further isotope to reduce
(W
, n

We next suppose that int int
= . We may assume that int int
1
= .
=

..
..

Figure 60.
=

..
. .

Figure 61.
Then there are two possibilities: int int
2
= and int int
2
= . In each
case, we can slide to along the disk
. Moreover, we can isotope and
slightly to reduce (W
, n
), a contradiction (cf. Figure 64 and Figure 65).


Let
1
and
2
(
1
and
2
resp.) be the components of (
resp.) with h
1

1
D
w
(h
1

1
D
w
resp.). Without loss of generality, we may suppose that
1

1
. Then
there are teo possibilities: (1)
2

2
and (2)
2

2
. In each case, we can slide

1
into
along the disk
. Moreover, we can isotope to reduce (W
, n
) is
reduced, a contradiction (cf. Figure 66 and Figure 67).
Let t
+
0
(t
0
resp.) be the rst critical height above t
0
(below t
0
resp.). Since
|P(t
0
) | = W
= max{w
(t)|t (1, 1)}, we see that the critical point of the

height t
+
0
(t
0
resp.) is a maximum or a -vertex (a minimum or a y-vertex resp.)
(see Figure 68).
=
1
D
w
w

..
.. ..

2
Figure 62.
=

..
..

1
..

2
Figure 63.
Lemma 3.3.9. The critical height t
0
is a y-vertex (not a minimum), or is
modied by edge slides so that the modied graph contains an unknotted cycle.
Proof. Suppose that the critical point of the height t
0
is a minimum. Let t
+
0
be a
regular height just above t
0
. Then (t
+
0
) contains a fat-vertex with a lower simple
outermost edge for the fat-vertex of (t
+
0
). Hence it follows from Lemma 3.3.8
that every simple outermost edge for each fat-vertex of (t
+
0
) is lower. Similarly,
every simple outermost edge for each fat-vertex of (t
0
) is upper. We now vary t
for t
+
0
to t
0
. Note that for each regular height t, all the simple outermost edges
for each fat-vertex of (t) are either upper or lower (Lemma 3.3.8); such a regular
height t is said to be upper or lower respectively. In these words, t
+
0
is lower and
t
0
is upper.
Let c
1
, . . . , c
n
(c
1
< < c
n
) be the critical heights of h|
2
contained in [t
+
0
, t
0
].
Note that the property upper or lower is unchanged at any height of [t
+
0
, t
0
] \
{c
1
, . . . , c
n
}. Hence there exists a critical height c
i
such that a height t is changed
from lower to upper at c
i
. The graph (t) is changed as in Figure 69 around the
critical height c
i
.
=

..
..

1
..

2
Figure 64.
=

..
..

1
..

2
Figure 65.
Let c
+
i
(c
i
resp.) be a regular height just above (below resp.) c
i
. We note that
the lower disk for (c
i
) and the upper disk for (c
+
i
) in Figure 69 are contained
in the same component of
2
, say D. We take parallel copies, say D
and D
, of
D such that D
is obtained by pushing D into one side and that D
is obtained by
pushing D into the other side (cf. Figure 70). Then we may suppose that there is
an upper (a lower resp.) simple outermost edge for a fat-vertex in D
(D
resp.).
Hence we can apply the arguments of the proof of Lemma 3.3.7 to modify so
that the modied graph contains an unknotted cycle.
Let v
be the y-vertex of at the height t
0
and t
0
a regular height just below
t
0
. Let v
be the intersection point of the descending edges from v
in and
P(t
0
), and let D
v
be the fat-vertex of (t
0
) correponding to v
.
Lemma 3.3.10. Every simple outermost edge for any fat-vertex of (t
0
) is lower,
or is modied by edge slides so that the modied graph contains an unknotted
cycle.
=

1
..

2
..
..

1
..

2
Figure 66.
=

1
..

2
..
..

1
..

2
Figure 67.
t
0
t
0
t
+
0
Figure 68.
w
of (t
0
) such that (t
0
) contains
an upper simple outermost edge for D
w
. Let be the edge of with h
1
D
w
.
Let

w
, ). Let (
resp.) be a union of the

components obtained by cutting by the two fat-vertices of (t
0
) incident to
upper
lower
w
t = c
+
i
t = c
i
t = c
i
Figure 69.
upper
w
lower
Figure 70.
(
resp.) such that a 1-handle correponding to each component intersects
\
(
resp.). Then (
resp.) satises one of the conditions (1), (2) and (3)

in the proof of Lemma 3.3.7.
Case A. D
v
= D
w
.
Then we have the following three cases. In each case, we can slide (a component
of) to along the disk
, n
), a
contradiction.
Case A-(1). satises the condition (1).
Then there are two possibilities: (i) v
and (ii) v
. In each case, see

Figure 71.
See Figure 72.
Then there are two possibilities: (i) v
and (ii) v
. In each case, see

Figure 73.
Case B. D
v
= D
w
.
Since
is upper, we see that does not satisfy the condition (2). Hence we
have the following.
(i) v

(ii) v

t = t
0
t = t
0
t = t
0
t = t
0
=
=
D
v
D
w

D
v
D
w
Figure 71.
t = t
0
t = t
0
=
D
v
D
w
Figure 72.
Case B-(1). satises the condition (1).
Since is upper, we see that the y-vertex of at the height t
0
is an endpoint
of , i.e., is the short vertical arc joining v
to v
. Then we can slide to

along the disk
to obtain a new graph
. Note that (W
, n
) = (W
, n
) (cf.
Figure 74). However, the critical point for
corresponding to v
is a minimum.
Hence we can apply the arguments in the proof of Lemma 3.3.9 to show that there
is an unknotted cycle in
.
t = t
0
t = t
0
=
D
v
D
w
Figure 73.
t = t
0
t = t
0
=
Figure 74.
Case B-(3). satises the condition (3).
Let
1
and
2
be the components of with
1
v
. Then we can slide

2
into
along the disk
, n
), a contradiction
(cf. Figure 75).
t = t
0
t = t
0
=
Figure 75.
Lemma 3.3.11. Every simple outermost edge for any fat-vertex of (t
0
) is in-
cident to D
v
, or is modied so that there is an unknotted cycle.
Proof. Suppose that (t
0
) contains a simple outermost edge for D
w
and is not
incident to D
v
. Then it follows from Lemma 3.3.10 that is lower. This means
that (t
0
) contains a lower simple edge, because an edge disjoint from D
v
is not
aected at all in [t
0
, t
0
]. This contradicts Lemma 3.3.8.
We now prove Proposition 3.3.4.

Proof of Proposition 3.3.4. We rst prove the following.
Claim. For any fat-vertex D
w
(= D
v
) of (t
0
), there are no loops of (t
0
)
based on D
w
, or is modied by edge slides so that the modied graph contains
an unknotted cycle.
w
(= D
v
) of (t
0
) such that there
is a loop of (t
0
) based on D
w
. Then separates cl(P(t
0
)\D
w
) into two disks
E
1
and E
2
with D
v
E
2
. By retaking D
w
and , if necessary, we may suppose
that there are no loop components of (t
0
) in int(E
1
). It follows from Lemma
3.3.5 that there is a fat-vertex D
w
of (t
0
) in int(E
1
). Then every outermost
edge for D
w
of (t
0
) is simple. Hence it follows from Lemma 3.3.11 that
contains an unknotted cycle and therefore we have the claim.
Then we have the following cases.
Case A. The descending edges of from the maximum or -vertex v
+
at the
height t
+
0
are equal to the ascending edges from v
(cf. Figure 76).

Figure 76.
Then we can immediately see that there is an unknotted cycle .
Case B. Exactly one of the descending edges from v
+
is equal to one of the
ascending edges from v
(cf. Figure 77).

Let
be the other edge disjoint from v
, and let w
be the rst intersection

point of P(t
0
) and the edge
. Let be an outermost edge for D

w
of (t
0
).
By the claim above, we see that is simple. It follows from Lemma 3.3.10 that
Figure 77.
we may suppose that is lower. It also follows from Lemma 3.3.11 that we may
suppose that the endpoints of are v
and w
. Let
be the outermost disk for

(D
w
, ). Set =
. Since the subarc of
whose endpoints are v
and w
is monotonous and is lower, we see that cannot satisfy the condition (3) in the
proof of Lemma 3.3.8. Hence satises the condition (1) or (2). In each case, we
to obtain a new graph with an unknotted cycle.

Case C. Any descending edge of from v
+
is disjoint from an ascending edge
from v
.
It follows from 3.3.10, 3.3.11 and the claim that (t
) contains a lower simple

outermost edge
i
(i = 1, 2) for D
w
i
which is adjacent to D
w
i
and D
v
. Let
i
w
i
,
i
). Set
i
=
i
i
. Since the subarc of
i
whose
endpoints v
+
and w
i
are monotonous and
i
is lower, we see that
i
cannot satisfy
the condition (3). Then we have the following.
Case C-(1). Both
1
and
2
If
1
=
2
, then we can slide
1
to
1
along the disk
1
. We can further isotope
to reduce (W
, n
), a contradiction (cf. Figure 78). Hence

1
=
2
.
Then we can slide
1

2
to
1
2
along
2
so that a new graph contains
an unknotted cycle (cf. Figure 79).
Case C-(2). Either
1
or
2
, say
1
, satises the condition (2).
Since
1
satises the condition (2), we see that the endpoints of
1
are v
+
and
v
. Hence w
2

1
. This implies that
2
satises the condition (1). Then we rst
slide
2
to
2
along
2
. We can further slide
1
to
1
along
1
so that a new graph
contains an unknotted cycle.
This completes the proof of Proposition 3.3.4.
=
Figure 78.
Figure 79.
3.4. Applications of Hakens theorem and Waldhausens theorem.
Corollary 3.4.1. Let M be a compact 3-manifold and (C
1
, C
2
; S) a reducible Hee-
gaard splitting. Then M is reducible or (C
1
, C
2
; S) is stabilized.
Proof. Suppose that M is irreducible. Let P be a 2-sphere such that P S is an
essential loop. Since M is irreducible, we see that P bounds a 3-ball in M. Hence
we can regard M as a connected sum of S
3
and M. By Theorem 3.3.1, the induced
Heegaard splitting of S
3
is stabilized. Hence this cancelling pair of disks shows
that (C
1
, C
2
; S) is stabilized.
Corollary 3.4.2. Any Heegaard splitting of a handlebody is standard, i.e, is ob-
tained from a trivial splitting by stabilization.
Exercise 3.4.3. Show Corollary 3.4.2.
Theorem 3.4.4. Let M be a closed 3-manifold. Let (C
1
, C
2
; S) and (C
1
, C
2
; S
)
be Heegaard splittings of M. Then there is a Heegaard splitting which is obtained
by stabilization of both (C
1
, C
2
; S) and (C
1
, C
2
; S
).
Proof. Let
C
1
and
C
1
be spines of C
1
and C
1
respectively. By an isotopy, we may
assume that
C
1
1
= and C
1
C
1
= . Set M
= cl(M\(C
1
C
1
)),
1
M
= C
1
and
2
M
= C
2
. Let (

C
1
,

C
2
;

S) be a Heegaard splitting of (M
;
1
M
,
2
M
). Set
C
1
= C
1
C
1
and C
2
= C
2
C
2
. Then it is easy to see that (C
1
, C
2
;

S) is a Heegaard
splitting of M. Note that C
2
= C
1
M
= C
1
(

C
1

C
2
) = (C
1

C
1
)

C
2
. Here,
we note that (C
1
,

C
2
;

S) is a Heegaard splitting of C
2
. It follows from Corollary
3.4.2 that (C
1
,

C
2
;

S) is obtained from a trivial splitting of C
2
by stabilization.
This implies that (C
1
, C
2
;

S) is obtained from (C
1
, C
2
; S
) by stabilization. On the
argument above, by replacing C
1
to C
1
, we see that (C
1
, C
2
;

S) is also obtained
from (C
1
, C
2
; S) by stabilization.
Remark 3.4.5. The stabilization problem is one of the most important themes on
Heegaard theory. But we do not give any more here. For the detail, for example,
see [8], [12], [15], [19] and [20].
4. Generalized Heegaard splittings
4.1. Denitions.
Denition 4.1.1. A 0-fork is a connected 1-complex obtained by joining a point
p to a point g whose 1-simplices are oriented toward g and away from p. For
n 1, an n-fork is a connected 1-complex obtained by joining a point p to each
of distinct n points t
i
(i = 1, ..., n) and to a point g whose 1-simplices are oriented
toward g and away from t
i
. We call p a root, t
i
a tine and g a grip.
Remark 4.1.2. An n-fork corresponds to a compression body C such that each
of t
i
(i = 1, 2, ..., n) corresponds to a component of
C and g correponds to
+
C
(cf. Figure 80).
tine
root
grip
Figure 80.
Denition 4.1.3. Let A (B resp.) be a collection of nite forks, T
A
(T
B
resp.)
a collection of tines of A (B resp.) and G
A
(G
B
resp.) a collection of grips of A
(B resp.). We suppose that there are bijections T : T
A
T
B
and G : G
A
G
B
.
A fork complex F is an oriented connected 1-complex A (B)/{T , G}, where
B denotes the 1-complex obtained by taking the opposite orientation of each
1-simplex and the equivalence relation /{T , G} is given by t T (t) for any t T
A
and g G(g) for any g G
A
. We dene:
1
F = {(tines of A) \ T
A
} {(grips of B) \ G
B
} and
2
F = {(tines of B) \ T
B
} {(grips of A) \ G
A
}.
Denition 4.1.4. A fork complex is exact if there exists e Hom(C
0
(F), R) such
that
(1) e(v
1
) = 0 for any v
1

1
F,
(2) (e)(e
A
) > 0 for any 1-simplex e
A
in A with the standard orientation,
(e)(e
A
) < 0 for any 1-simplex e
B
in B with the standard orientation, where
denotes the coboundary operator Hom(C
0
(F), R) Hom(C
1
(F), R)
and
(3) e(v
2
) = 1 for any v
2

2
F.
Remark 4.1.5. Geometrically speaking, F is exact if and only if we can put F
in R
3
so that
(1)
1
F lies in the plane of height 0,
(2) for any path in F from a point in
1
F to a point in
2
F, h|
is mono-
tonically increasing, where h is the height function of R
3
and
(3)
2
F lies in the plane of height 1 (cf. Figure 81).
1
0
Figure 81.
In the following, we regard fork complexes as geometric objects, i.e., 1-dimensional
polyhedra.
Denition 4.1.6. A fork of F is the image of a fork in AB in F. A grip (root
and tine resp.) of F is the image of a grip (root and tine resp.) in A B in F.
Denition 4.1.7. Let M be a compact orientable 3-manifold, and let (
1
M,
2
M)
be a partition of boundary components of M. A generalized Heegaard splitting
of (M;
1
M,
2
M) is a pair of an exact fork complex F and a proper map :
(M;
1
M,
2
M) (F;
1
F,
2
F) which satises the following.
(1) The map is transverse to F {the roots of F}.
(2) For each fork F F, we have the following (cf. Figure 82).
(a) If F is a 0-fork, then
1
(F) is a handlebody V
F
such that (1)
1
(g) =
V
F
and (2)
1
(p) is a 1-complex which is a spine of V
F
, where g is
the grip of F.
(b) If F is an n-fork with n 1, then
1
(F) is a connected compression
body V
F
such that (1)
1
(g) =
+
V
F
, (2) for each tine t
i
,
1
(t
i
) is a
connected component of
V
F
and
1
(t
i
) =
1
(t
j
) for i = j and (3)
1
(p) is a 1-complex which is a deformation retract of V
F
, where g is
the grip of F, p is the root of F and {t
i
}
1in
is the set of the tines
of F.
Remark 4.1.8. Let g be a grip of F which is contained in the interior of F. Let
F
1
and F
2
be the forks of F which are adjacent to g. Then (
1
(F
1
),
1
(F
2
);
1
(g))
is a Heegaard splitting of
1
(F
1
F
2
).
Figure 82.
Denition 4.1.9. A generalized Heegaard splitting (F, ) is said to be strongly
irreducible if (1) for each tine t,
1
(t) is incompressible, and (2) for each grip g
with two forks attached to g, say F
1
and F
2
, (
1
(F
1
),
1
(F
2
);
1
(g)) is strongly
irreducible.
Let M be the set of nite multisets of Z
0
= {0, 1, 2, ...}. We dene a total
order < on M as follows. For M
1
and M
2
M, we rst arrange the elements of
M
i
(i = 1, 2) in non-increasing order respectively. Then we compare the arranged
tuples of non-negative integers by lexicographic order.
Example 4.1.10. (1) If M
1
= {5, 4, 1, 1} and M
2
= {5, 3, 2, 2, 2, 1}, then M
2
<
M
1
.
(2) If M
1
= {3, 1, 0, 0} and M
2
= {3, 1, 0, 0, 0}, then M
1
< M
2
.
Denition 4.1.11. Let (F, ) be a generalized Heegaard splitting of (M;
1
M,
2
M).
We dene the width of (F, ) to be the multiset
w(F, ) = {genus(
1
(g
1
)), . . . , genus(
1
(g
m
))},
where {g
1
, . . . , g
m
} is the set of the grips of F. We say that (F, ) is thin if
w(F, ) is minimal among all generalized Heegaard splittings of (M;
1
M,
2
M).
Example 4.1.12. The thin generalized Heegaard splittings of the 3-ball B
3
are two
fork complexes illustrated in Figure 83, where
1
(F
1
) is a 3-ball and
1
(F
2
)

=
S
2
[0, 1].
F
1
F
1
F
2
Figure 83.
4.2. Properties of thin generalized Heegaard splittings. In this subsection,
let (F, ) be a thin generalized Heegaard splitting of (M;
1
M,
2
M).
Observation 4.2.1. Let t be a tine of F. Then any 2-sphere component of
1
(t)
is essential in M unless M is a 3-ball.
Proof. Suppose that there is a tine t such that
1
(t) is a 2-sphere, say P, which
bounds a 3-ball B in M. Let F
B
be the subcomplex of F with
1
(F
B
) = B.
If F
B
= F, then we see that M is a 3-ball. Otherwise, there is a fork F
with
t F
and F
F
B
. Let e
t
be the 1-simplex in F
joining t to the root of F
. Set
F
= F \ (F
B
e
t
). Note that
1
(F
F
B
) (=
1
(F
) B) is a compression
body V

. Then it is easy to see that we can modify in V

to obtain
: M F
such that (
)
1
(F
\ e
t
) is the compression body V

(cf. Figure 84).
t

Figure 84.
Moreover, the generalized Heegaard structure on (F, ) (e.g. A, B decompo-
sition etc) is naturally inherited to (F
). Then we clearly have w(F
) <
w(F, ), contradicting the assumption that (F, ) is thin.
Lemma 4.2.2. Suppose that there is a fork F such that

1
(t) is trivial. Let t be
the tine of F. Then
1
(t) is a component of M and one of the following holds.
(1) M is a 3-ball.
(2) M

=
1
(t) [0, 1].
(3)
1
(t) is compressible in M.
Proof. We rst prove that
1
(t) is a boundary component of M. Suppose that
1
(t) is not a boundary component of M. Let g be the grip of F. If
1
(g) is
a boundary component of M, then we can reduce the width by removing F, a
=
1 1
F
Figure 85.
Hence F is contained in the interior of F. Note that F is a 1-fork. Let F
1
be
the fork attached to g and F
2
the fork attached to t. Note that since
1
(F) is
a trivial compression body, we see that
1
(F
1
F F
2
) is also a compression
body. Hence we can replace F
1
F F
2
in F to a new fork so that we can
obtain a new fork complex, say F
. Moreover, we can modify : M F

to obtain
: M F
so that (F
) is a generalized Heegaard splitting of

(M;
1
M,
2
M) with w(F
) < w(F, ), a contradiction (cf. Figure 86). Hence
1
(t) is a boundary component of M.
=
F
1
F
F
2
Figure 86.
We next show that one of the conclusions (1)-(3) of Lemma 4.2.2 holds. Suppose
that both conclusions (1) and (2) of Lemma 4.2.2 do not hold, i.e., M is not a 3-ball
and M
=
1
(t) [0, 1]. Then there is a fork F
(= F) attached to g. Moreover,
since (F, ) is thin and M
=
1
(t) [0, 1], we see that
1
(F
) is a non-trivial
compression body. Also, since M is not a 3-ball,
1
(t) is not a 2-sphere. Hence
we see that
1
(t) is compressible in
1
(F)
1
(F
). This implies that the

conclusion (3) of Lemma 4.2.2 holds.
Proposition 4.2.3. Let F
1
and F
2
be forks of F which have the same grip g of
F. Then (
1
(F
1
),
1
(F
2
);
1
(g)) is strongly irreducible.
Proof. Set A
g
=
1
(F
1
), B
g
=
1
(F
2
), S
g
=
1
(g), M
g
= A
g
B
g
,
1
M
g
=
A
g
and
2
M
g
=
B
g
. Then (A
g
, B
g
; S
g
) is a Heegaard splitting of (M
g
;
1
M
g
,
2
M
g
).
Suppose that (A
g
, B
g
; S
g
) is weakly reducible. Let D
A
and D
B
be meridian disks
of A
g
and B
g
respectively which satisfy D
A
D
B
= . Let
A
(
B
resp.)
be a complete meridian system of A
g
(B
g
resp.) such that D
A
(D
B
resp.) is a
component of
A
(
B
resp.) (cf. (6) of Remark 3.1.3). Note that A
g
is obtained
from
A
g
[0, 1] and 0-handles H
0
1
corresponding to
A
(cf. (3) of Remark 3.1.3) and that B
g
is obtained from S
g
[0, 1] by attaching
2-handles H
2
corresponding to
B
and 3-handles H
3
(cf. Denition 3.1.1). Hence
we see that M
g
admits the following decomposition (cf. Remark 3.1.13):
M
g
= (
1
M
g
[0, 1]) H
0
H
1
H
2
H
3
.
Let h
1
be the component of H
1
corresponding to D
A
and h
2
the component of H
2
corresponding to D
B
. Then M
g
admits the following decomposition:
M
g
= (
1
M
g
[0, 1]) H
0
(H
1
\ h
1
) h
2
h
1
(H
2
\ h
2
) H
3
.
Set A
g
= (
1
M
g
[0, 1]) H
0
(H
1
\ h
1
). We divide the proof into the following
two cases.
Case 1. D
A
or D
B
is non-separating in S
g
.
Suppose rst that D
A
g
. Then A
g
is a compression body
(cf. (6) of Remark 3.1.3). Since A
g
= (
1
M
g
[0, 1]) H
0
(H
1
\ h
1
), we obtain
M
g
= A
g
h
2
h
1
(H
2
\ h
2
) H
3
.
Note that the attaching region of the 2-handle h
2
is contained in
+
A
g
. Hence we
have:
M
g

= A
g

_
(
+
A
g
[0, 1]) h
2
_
h
1
(H
2
\ h
2
) H
3
.
Set B
g
= (
+
A
g
[0, 1]) h
2
. Then B
g
is also a compression body and we have:
M
g

= A
g
B
g
h
1
(H
2
\ h
2
) H
3
.
Note that
g
is homeomorphic to the surface obtained from S
g
by performing
surgery along D
A
D
B
. Then we have the following subcases.
Case 1.1. D
B
g
and D
A
D
B
g
.
Then
g
is connected. Note that
M
g

= A
g
B
g
h
1
(H
2
\ h
2
) H
3
= A
g
B
_
(
g
[0, 1]) h
1
_
(H
2
\ h
2
) H
3
.
Set A
g
= (
g
[0, 1]) h
1
. Then A
g
is also a compression body and we have:
M
g

= A
g
B
g
A
g
(H
2
\ h
2
) H
3
= A
g
B
g
A
g

_
(
+
A
g
[0, 1] (H
2
\ h
2
) H
3
_
.
Set B
g
=
+
A
g
[0, 1] (H
2
\ h
2
) H
3
. Note that B
g
A
g
=
g
=
g
. This
shows that each handle of H
2
\ h
2
and H
3
is adjacent to A
g
along
+
A
g
. This
implies that B
g
is also a compression body. Hence we have:
M
g

= (A
g
B
g
) (A
g
B
g
).
Then we can substitute F
1
F
2
in F for F
1
F
2
F
1
F
2
, where F
1
, F
2
, F
1
and F
2
are forks corresponding to A
g
, B
g
, A
g
and B
g
respectively. Set F
=
(F \ (F
1
F
2
)) (F
1
F
2
F
1
F
2
). Then we can modify : M F
in M
g
to obtain
: M F
such that (
)
1
(F
1
) = A
g
, (
)
1
(F
2
) = B
g
,
(
)
1
(F
1
) = A
g
and (
)
1
(F
2
) = B
g
. It is easy to see that w(F
) < w(F, ),
a contradiction (cf. Figure 87).
Case 1.2. D
B
g
and D
A
D
B
is separating in S
g
.
Then
g
consists of two components, say G
1
and G
2
.
D
A
A
g
D
B
B
g

A
g
B
g
A
g
B
g
Figure 87.
M
g

= A
g
B
g
h
1
(H
2
\ h
2
) H
3
= A
g
B
g
(
g
[0, 1]) h
1
(H
2
\ h
2
) H
3
= A
g
B
g
((G
1
G
2
) [0, 1]) h
1
(H
2
\ h
2
) H
3
.
Set A
g
= ((G
1
G
2
) [0, 1]) h
1
. Since D
B
g
, we see that
h
1
joins G
1
to G
2
. Hence A
g
is a compression body and we have:
M
g

= A
g
B
g
A
g
(H
2
\ h
2
) H
3
= A
g
B
g
A
g
(
+
A
g
[0, 1]) (H
2
\ h
2
) H
3
.
Set B
g
=
+
A
g
[0, 1] (H
2
\ h
2
) H
3
. Note that B
g
A
g
=
g
=
g
. This
shows that each handle of H
2
\ h
2
and H
3
is adjacent to A
g
along
+
A
g
. This
implies that B
g
is also a compression body. Hence we have:
M
g

= (A
g
B
g
) (A
g
B
g
).
According to this decomposition, we can modify the fork complex (F, ) as in
Figure 88 or Figure 89. It is easy to see that for a new complex (F
), we have
w(F
) < w(F, ), a contradiction.

Case 1.3. D
B
is separating in S
g
(hence D
A
D
B
is separating in S
g
).
Then
g
consists of two components, say

G
1
and

G
2
. Since D
B
is separating
in S
g
, we see that h
1
joins

G
1
or

G
2
, say

G
1
, to itself. Let H
2
1
(H
2
2
resp.) be the
components of H
2
\ h
2
adjacent to

G
1
(

G
2
resp.). Let H
3
1
(H
3
2
resp.) be the
components of H
3
adjacent to

G
1
(

G
2
resp.). Set

B
g
= B
g
H
2
2
H
3
2
. Then
D
A
A
g
D
B
B
g

A
g
B
g
A
g
B
g
Figure 88. The case of irreducible splittings
D
A
A
g
D
B
B
g

A
g
B
g
A
g
B
g
Figure 89. The case of reducible splittings
g
+

B
g
=
+
B
g
. Set A
g
= (

G
1
[0, 1]) h
1
and
B
g
= (
+
A
g
[0, 1]) H
2
1
H
3
1
. Then each of A
g
and B
g
is a compression body.
Hence we have:
M
g

= A
g
B
g
h
1
(H
2
1
H
2
2
) (H
3
1
H
3
2
)
= A
g
(B
g
H
2
2
H
3
2
) h
1
H
2
1
H
3
1
= A
g

B
g
(

G
1
[0, 1]) h
1
H
2
1
H
3
1
= A
g

B
g
A
g
(
+
A
g
[0, 1]) H
2
1
H
3
1
= (A
g
) (A
g
B
g
).
Figure 90. It is easy to see that for a new complex (F
), we have w(F
) <
w(F, ), a contradiction. Therefore if D
A
is non-separating, we have the desired
conclusion.
D
A
A
g
D
B
B
g

A
g
A
g
B
g
Figure 90.
Suppose next that D
B
g
. Then we start with the dual
handle decomposition
M
g
= (
2
M
g
[0, 1])

H
0

H
1

H
2

H
3
and apply the above arguments which gives a contradiction.
Case 2. Each of D
A
and D
B
is separating in S.
Then A
g
consists of two compression bodies, say

A
g
and

A
g
(cf. (6) of Remark
3.1.3). We may suppose that h
2
is attached to
+

A
g
. Set B
g
= (
+

A
g
[0, 1])h
2
.
Since D
B
is separating in S, we see that
g
consists of two components, say
G
1
and G
2
. Note that D
A
is also separating in S
g
. Hence we may suppose that
h
1
G
2
= and h
1
G
1
= . Let H
2
1
be the components of H
2
\ h
2
adjacent to
G
1
and H
3
1
be the components of H
3
adjacent to G
1
. Set H
2
2
= H
2
\ (h
2
H
2
1
),
H
3
2
= H
3
\ H
3
1
and

B
g
= B
g
H
2
1
H
3
1
. Then

B
g
is a compression body. Set
A
g
= (G
2
[0, 1])

A
g
h
1
and B
g
= (
+
A
g
[0, 1]) H
2
2
H
3
2
. Note that each
of A
g
and B
g
is a compression body. Set A
g
=

A
g
A
g
. Note also that A
g
is a
compression body (cf. (5) of Remark 3.1.3). Hence we have:
M
g

= A
g
h
2
h
1
(H
2
1
H
2
2
) (H
3
1
H
3
2
)
= (

A
g

A
g
) h
2
h
1
(H
2
1
H
2
2
) (H
3
1
H
3
2
)
=

A
g

_
(
+

A
g
[0, 1]) h
2
_
g
h
1
(H
2
1
H
2
2
) (H
3
1
H
3
2
)
=

A
g
B
g
((G
1
G
2
) [0, 1])

A
g
h
1
(H
2
1
H
2
2
) (H
3
1
H
3
2
)
=

A
g
B
g
((G
1
G
2
) [0, 1])

A
g
h
1
(H
2
1
H
2
2
) (H
3
1
H
3
2
)
=

A
g

_
B
g
(G
1
[0, 1]) H
2
1
H
3
1
_
_
(G
2
[0, 1])

A
g
h
1
_
H
2
2
H
3
2
=

A
g

B
g
(

A
g
A
g
)
_
(
+
A
g
[0, 1]) H
2
2
H
3
2
_
= (

A
g

B
g
) (A
g
B
g
).
Figure 91 or Figure 92. It is easy to see that for a new complex (F
), we have
w(F
) < w(F, ), a contradiction.
Lemma 4.2.4. Any component

1
(t) is incompressible in M unless M is -
compressible, where t is a tine of F.
Proof. Suppose that
1
(t) is compressible in M for a tine t of F. Let D be a
compressing disk of
1
(t). Let T be the union of the tines of F. By an in-
nermost disk argument, we may assume that D
1
(T ) = D. Let F
1
be the
fork containing ((D; D)). Note that
1
(t) is incompressible in
1
(F
1
) (cf.
(4) of Remark 3.1.3). Hence there is a fork F
2
(= F
1
) attached to the grip, say
g, of F
1
. Since D
1
(T ) = D, we have D
1
(F
1
F
2
). Hence D is a
-compressing disk of M
=
1
(F
1
)
1
(F
2
). Hence it follows from (2) of Theo-
rem 3.2.1 and Lemma 3.1.20 that the Heegaard splitting (
1
(F
1
),
1
(F
2
);
1
(g))
is either weakly reducible or trivial. It also follows from Proposition 4.2.3 that
(
1
(F
1
),
1
(F
2
);
1
(g)) is strongly irreducible and hence the splitting must be
trivial. Since
1
(F
1
) contains (D; D), we see that
1
(F
1
) is a trivial compres-
sion body and that t is the tine of F
1
. Hence by Lemma 4.2.2, we have one of the
following: (1) M is a 3-ball, (2) M

=
1
(t) [0, 1] and (3) M is -compressible.
We suppose that M does not satisfy the condition (3), i.e., M is -incompressible.
If M satises the condition (1), i.e., M is a 3-ball, then it follows from Exam-
ple 4.1.12 that t is the only tine of F and that
1
(t) is incompressible in M.
This contradicts that we suppose that
1
(t) is compressible in M. If M satises
the condition (2), i.e., M

=
1
(t) [0, 1], then
1
(t) is incompressible in M, a
contradiction.
D
A
A
g
D
B
B
g
g

k
l
g k
g l
g
Figure 91. The case of irreducible splittings
As a direct consequence of Proposition 4.2.3 and Lemma 4.2.4, we have the
following.
Corollary 4.2.5. (F, ) is strongly irreducible unless M is -compressible.
Remark 4.2.6. There are strongly irreducible splittings which are not thin. In
fact, there are strongly irreducible Heegaard splittings which are not minimal genus
(cf. [2] and [7]).
Lemma 4.2.7. Suppose that F contains a tine. There exists a tine t of F such
that
1
(t) is a 2-sphere if and only if M is reducible or is a 3-ball.
Proof. The only if part is immediate from Observation 4.2.1. Hence we will give
a proof of the if part.
Suppose that M is reducible or is a 3-ball. If M is a 3-ball, then it follows from
Example 4.1.12 that there is exactly one tine, say t, of F and
1
(t) = M is a
2-sphere. Hence in the remainder of the proof, we suppose that M is reducible.
Let T be the union of the tines of F. Let P be a reducing 2-sphere such that
|P
1
(T )| is minimal among such all reducing 2-spheres. By an innermost disk
argument, we see that P
1
(T ) = . Let F
1
be a fork of F with
1
(F
1
)P = .
Suppose rst that there are no forks of F attaching to the grip of F
1
. Then
this implies that P is an essential 2-sphere in
1
(t).
Suppose next that there is a fork of F, say F
2
, other than F
1
which attaches to
the grip, say g, of F
1
. Note that
1
(F
1
)
1
(F
2
) contains P. It follows from (1) of
Theorem 3.2.1 that
1
(F
1
) or
1
(F
2
) is reducible, or (
1
(F
1
),
1
(F
2
);
1
(g))
D
A
A
g
D
B
B
g
g

g
Figure 92. The case of reducible splittings
is reducible. The latter condition, however, contradicts Proposition 4.2.3. Hence
we may assume that
1
(F
1
) is reducible, that is, there is a 2-sphere component
P
0
of
(
1
(F
1
)) (cf. (1) of Remark 3.1.3). This implies that there is a tine t
with
1
(t) = P
0
.
Lemma 4.2.8. If some
1
(g) is a torus, where g is a grip of F, then one of the
following holds.
(1) M is reducible.
(2) M is (a torus) [0, 1].
(3) M is a solid torus.
(4) M is a lens space.
Proof. Suppose that M does not satisfy the conclusion (1) of Lemma 4.2.8, i.e.,
M is irreducible. Note that
1
(g) may be a boundary component of M. Let F
be a fork such that the grip of F is g. Set V =
1
(F).
If V is trivial, then M is either T
2
[0, 1] or a solid torus by Lemma 4.2.2.
Hence conclusion (2) or (3) of Lemma 4.2.8 holds.
If V is non-trivial, then we see that V is a solid torus by Observation 4.2.1
and Example 4.1.12. Suppose further that the conclusion (3) does not hold.,
i.e., M is not a solid torus. Then there is a fork F
(= F) attached to g. Set
V

=
1
(F
). If V

is trivial, then it follows from Lemma 4.2.2 that M is a solid
torus, a contradiction. If V

is non-trivial, then we see that V

is a solid torus by
Observation 4.2.1 and Example 4.1.12. Hence M is a lens space and we have the
conclusion (4) of Lemma 4.2.8.
4.3. Examples of generalized Heegaard splittings. In this section, we use
some theorems without proofs to obtain generalized Heegaard splittings and as-
sociated fork complexes. Let F
g
be a connected closed orientable surface of genus
g.
M = F
g
[0, 1].
Set M = F
g
[0, 1], A = F
g
[0, 1/2], B = F
g
[1/2, 1] and S = F
g
{1/2}.
Clearly, (A, B; S) is a Heegaard splitting of M, and we call this Heegaard splitting
the trivial Heegaard splitting of type I. Let p be a point in F
g
. Set
A
= ((F
g
{0}) (p [0, 1]) (F
g
{1}); M),
B
= cl(M \ A
) and S
= A
. Then (A
, B
; S
) is also a Heegaard splitting of

M, and we call this splitting the trivial Heegaard splitting of type II. The proof of
the next observation is left to the reader.
Observation 4.3.1. Both these Heegaard splittings are strongly irreducible.
In fact, Scharlemann and Thompson proved the following.
Theorem 4.3.2 ([16] 2.11 Main Theorem). Any irreducible Heegaard splitting of
F
g
[0, 1] is trivial of type I or II.
We remark that the fork complexes associated to these Heegaard splittings are
illustrated in Figure 93.
Type I
Type II
Figure 93.
M = F
g
S
1
.
Note that S
1
is regarded as [0, 1]/{0} {1}. Let p and q be distinct points in
F
g
. Set
A = cl((F
g
[0, 1/2]) \ (p [0, 1/2]; F
g
[0, 1/2])) (q [1/2, 1]; F
g
[1/2, 1])
and
B = cl(M \ A)
= cl((F
g
[1/2, 1]) \ (q [1/2, 1]; F
g
[1/2, 1])) (p [0, 1/2]; F
g
[0, 1/2]).
Note that A and B are handlebodies. Set S = A B. Then (A, B; S) is a
Heegaard splitting of M = F
g
S
1
and is called the trivial Heegaard splitting of
M = F
g
S
1
(cf. Figure 94).
Exercise 4.3.3. Show that this trivial Heegaard splitting is weakly reducible.
p
q q p
A
B
Figure 94.
Theorem 4.3.4 ([18] Theorem 5.7). Any irreducible Heegaard splitting of F
g
S
1
is the trivial splitting.
T
3
= T
2
S
1
.
I
1
I
2
p
q
A
B
(a) (b)
Figure 95.
It is known that T
3
is obtained from a cube [0, 1] [0, 1] [0, 1] by attaching
corresponding edges and faces as in Figure 95 (a). Set
A = cl((T
2
[0, 1/2]) \ (p [0, 1/2]; T
2
[0, 1/2])) (q [1/2, 1]; T
2
[1/2, 1])
and
B = cl(T
3
\ A)
= cl((T
2
[1/2, 1]) \ (q [1/2, 1]; T
2
[1/2, 1])) (p [0, 1/2]; T
2
[0, 1/2]).
Then we see that A and B are genus two handlebodies and that it follows from
Theorem 4.3.4 that (A, B; S) is the Heegaard splitting of T
3
, where S = A = B
(cf. Figure 95 (b)). Set h
1
= (q[1/2, 1]; T
2
[1/2, 1]) and h
2
= (p[0, 1/2]; T
2
[0, 1/2]). Note that h

1
(h
2
resp.) can be regarded as a 1-handle (2-handle resp.) in
a handle decomposition of T
3
obtained from the Heegaard splitting (A, B; S). Since
h
1
h
2
= , we can perform a weak reduction to obtain a generalized Heegaard
splitting. We give a concrete description of the generalized Heegaard splitting in
the following. First, set A
1
= cl(T
3
[0, 1/2] \ h
1
) and B
2
= cl(T
3
[1/2, 1] \ h
2
).
That is, A
1
is obtained from A by removing the 1-handle h
1
and B
2
is obtained
from B by removing the 2-handle h
2
. Then we have:
T
3
= A B
= A
1
h
1
h
2
B
2
= A
1
(A
1
[0, 1]) h
1
h
2
B
2
= A
1
_
(A
1
[0, 1]) h
2
_
h
1
B
2
.
Set B
1
= (A
1
[0, 1]) h
2
. Then B
1
is a compression body such that
+
B
1
=
A
1
and
B consists of two tori. Hence we have:

T
3

= A
1
B
1
h
1
B
2
= A
1
B
1
_
(
B
1
[0, 1]) h
1
_
B
2
.
Set A
2
= (
B
1
[0, 1]) h
1
. Then A
2
+
A
2
=
B
2
and
A
2
=
B
1
. Hence we have:
T
3
= (A
1
B
1
) (A
2
B
2
).
This together with the fork complex as in Figure 96 gives a generalized Heegaard
splitting.
B
1
A
1
A
2
B
2
A
1
B
1
A
2
B
2
Figure 96.
Exercise 4.3.5. Show that this is the only fork complex associated with a gener-
alized Heegaard splitting of T
3
via weak reduction.
Remark 4.3.6. The inverse procedure of weak reduction is called amalgamation.
Exercise 4.3.7. Show that the above generalized Heegaard splitting of T
3
is
strongly irreducible.
M = F
2
S
1
.
a
b
a
d
c
d
c
b
a
b
a
d
c
d
c
b
B
A
B
1
A
1
A
2
B
2
(a) (b) (c)
Figure 97.
M is obtained from a (an octagon) [0, 1] by attaching corresponding edges and
faces as in Figure 97 (a). Set
A = cl((F
g
[0, 1/2]) \ (p [0, 1/2]; F
g
[0, 1/2])) (q [1/2, 1]; F
g
[1/2, 1])
and
B = cl(M \ A)
= cl((F
g
[1/2, 1]) \ (q [1/2, 1]; F
g
[1/2, 1])) (p [0, 1/2]; F
g
[0, 1/2]).
Then it follows from Theorem 4.3.4 that we obtain the Heegaard splitting M =
AB ( see Figure 97 (b)). As descibed in case of M = T
3
, we can perform a weak
reduction and we obtain the same fork complex as that illustrated in Figure 96.
In this case, each of A
1
and B
2
is a handlebody of genus four and each of A
2
and
B
1
+
A
2
= B
2
, A
1
=
+
B
1
and
A
2
=
B
1
.
For the Heegaard splitting A B of M, we can nd another weak reduction as
follows. Recall that M = P
8
[0, 1]/ , where P
8
is an octagon (cf. Figure 97).
Then there is a handle decomposition
M = h
0
h
1
a
h
1
b
h
1
c
h
1
d
h
1
e
h
2
a
h
2
b
h
2
c
h
2
d
h
2
e
h
3
,
where a 0-handle h
0
corrsponds to a vertex of P
8
, a 1-handle h
1
a
(h
1
b
, h
1
c
, h
1
d
and
h
1
e
resp.) corrsponds to a (b, c, d and e resp.) in P
8
, a 2-handle h
2
a
(h
2
b
, h
2
c
and
h
2
d
resp.) corrsponds to the face bounded by eae
1
a
1
(ebe
1
b
1
, ece
1
c
1
and
ede
1
d
1
resp.) in P
8
[0, 1], a 2-handle h
2
e
corrsponds to the face bounded by
aba
1
b
1
cdc
1
d
1
in P
8
and a 3-handle h
3
corrsponds to the vertex in the interior
of P
8
[0, 1]. Set A
1
= h
0
h
1
a
h
1
e
. Then A
1
is a genus two handlebody and we
have:
M = A
1
h
1
b
h
1
c
h
1
d
h
2
a
h
2
b
h
2
c
h
2
d
h
2
e
h
3
= A
1
(A
1
[0, 1]) h
1
b
h
1
c
h
1
d
h
2
a
h
2
b
h
2
c
h
2
d
h
2
e
h
3
= A
1
_
(A
1
[0, 1]) h
2
a
_
h
1
b
h
1
c
h
1
d
h
2
b
h
2
c
h
2
d
h
2
e
h
3
Set B
1
= (A
1
[0, 1])h
2
a
. Then B
1
+
B
1
= A
1
and
B
1
consists of two tori. Then we have:
M

= A
1
B
1
h
1
b
h
1
c
h
1
d
h
2
b
h
2
c
h
2
d
h
2
e
h
3
= A
1
B
1
(
B
1
[0, 1]) h
1
b
h
1
c
h
1
d
h
2
b
h
2
c
h
2
d
h
2
e
h
3
= A
1
B
1
_
(
B
1
[0, 1]) h
1
b
_
h
1
c
h
1
d
h
2
b
h
2
c
h
2
d
h
2
e
h
3
.
Set A
2
= (
B
1
[0, 1]) h
1
b
. Then A
2
+
A
2
is
a closed surface of genus two and
A
2
=
B
1
. Then we have:
M

= A
1
B
1
A
2
h
1
c
h
1
d
h
2
b
h
2
c
h
2
d
h
2
e
h
3
= A
1
B
1
A
2
(
+
A
2
[0, 1]) h
1
c
h
1
d
h
2
b
h
2
c
h
2
d
h
2
e
h
3
= A
1
B
1
A
2
_
(
+
A
2
[0, 1]) h
2
b
_
h
1
c
h
1
d
h
2
c
h
2
d
h
2
e
h
3
.
Set B
2
= (
+
A
2
[0, 1]) h
2
b
. Then B
2
+
B
2
=
+
A
2
and
B
2
consists of a torus. Then we have:
M

= A
1
B
1
A
2
B
2
h
1
c
h
1
d
h
2
c
h
2
d
h
2
e
h
3
= A
1
B
1
A
2
B
2
(
B
2
[0, 1]) h
1
c
h
1
d
h
2
c
h
2
d
h
2
e
h
3
= A
1
B
1
A
2
B
2
_
(
B
2
[0, 1]) h
1
c
_
h
1
d
h
2
c
h
2
d
h
2
e
h
3
.
Set A
3
= (
B
2
[0, 1]) h
1
c
. Then A
3
+
A
3
is
A
3
=
B
2
. Then we have:
M

= A
1
B
1
A
2
B
2
A
3
h
1
d
h
2
c
h
2
d
h
2
e
h
3
= A
1
B
1
A
2
B
2
A
3
(
+
A
3
[0, 1]) h
1
d
h
2
c
h
2
d
h
2
e
h
3
= A
1
B
1
A
2
B
2
A
3
_
(
+
A
3
[0, 1]) h
2
c
_
h
1
d
h
2
d
h
2
e
h
3
.
Set B
3
= (
+
A
3
[0, 1]) h
2
c
. Then B
3
+
B
3
=
+
A
3
and
B
3
consists of two tori. Then we have:
M

= A
1
B
1
A
2
B
2
A
3
B
3
h
1
d
h
2
d
h
2
e
h
3
= A
1
B
1
A
2
B
2
A
3
B
3
(
B
3
[0, 1]) h
1
d
h
2
d
h
2
e
h
3
= A
1
B
1
A
2
B
2
A
3
B
3
_
(
B
3
[0, 1]) h
1
d
_
h
2
d
h
2
e
h
3
.
Set A
4
= (
B
4
[0, 1]) h
1
d
. Then A
4
+
A
4
is
A
4
=
B
3
. Then we have:
M

= A
1
B
1
A
2
B
2
A
3
B
3
A
4
h
2
d
h
2
e
h
3
= A
1
B
1
A
2
B
2
A
3
B
3
A
4
(
+
A
4
[0, 1]) h
2
d
h
2
e
h
3
.
Set B
4
= (
+
A
4
[0, 1]) h
2
d
h
2
e
h
3
. Then B
4
is a genus two handlebody such
that B
4
=
+
A
4
. Therefore we have the following decomposition.
T
3
= (A
1
B
1
) (A
2
B
2
) (A
3
B
3
) (A
4
B
4
).
A
1
B
1
A
2
B
2
A
3
B
3
A
4
B
4
. .
(a torus with
a single hole)S
1
. .
(a torus with
a single hole)S
1
Figure 98.
This together with the fork complex as in Figure 98 gives a generalized Heegaard
splitting. We remark that (A
1
B
1
) (A
2
B
2
) ((A
3
B
3
) (A
4
B
4
) resp.)
composes a (a torus with a single hole) S
1
.
By changing the attaching order of h
1
b
, h
1
c
and h
1
d
, we can obtain two more
strongly irreducible generalized Heegaard splittings via weak reduction (cf. Figure
99).
A
1
B
1
A
3
B
3
A
2
B
2
A
4
B
4
A
1
B
1
A
3
B
3
A
2
B
2
A
4
B
4
Figure 99.
Exercise 4.3.8. Show that these are the only fork complexes associated with a
generalized Heegaard splitting of F
2
S
1
via weak reduction.
Remark 4.3.9. The fork complexes associated with distinct weak reduction of a
Heegaard splitting need not homotopic.
References
[1] A. Casson and C. McA. Gordon, Reducing Heegaard splittings, Topology Appl. 27 (1987),
275-283.
[2] A. Casson and C. McA. Gordon, Manifolds with irreducible Heegaard splittings of arbitarily
high genus (unpublished).
[3] D. Gabai, Foliations and the topology of 3-manifolds III, J. Dierential Geometry 26 (1987),
479-536.
[4] W. Haken, Some results on surfaces in 3-manifolds, Studies in Modern Topology, Math.
Assoc. Amer. 1968, 39-98.
[5] A. Hatcher and W. Thurston, A presentation for the mapping class group of a closed ori-
entable surface, Topology 19 (1980), 221-237.
[6] K. Johannson, On surfaces and Heegaard surfaces, Trans. Amer. Math. Soc. 325 (1991),
573-591.
[7] T. Kobayashi, A construction of 3-manifolds whose homeomorphism classes of Heegaard
splittings have polynomial growth, Osaka J. Math. 29 (1992), 653-674.
[8] F. Lei, On stability of Heegaard splittings, Math. Proc. Camb. Phil. Soc. 129 (2000), 55-57.
[9] J. Milnor, Morse theory, Ann. of Math. Studies 51, Princeton Univ. Press, Princeton, 1963.
[10] E. Moise, Ane structures in 3-manifolds whose homeomorphism classes of Heegaard split-
tings have polynomial growth, Ann. of Math. 55 (1952), 96-114.
[11] J. -P. Otal, Sur les scindements de Heegaard de la sphere S
3
, Topology 30 (1991), 249-258.
[12] K. Reidemeister, Zur dreidimensionalen Topologie, Abh. Math. Sem. Univ. Hamburg 9
(1933), 189-194.
[13] D. Rolfsen, knots and links, Math. Lect. Note Series 7. Publish or Perish Inc., 1976.
[14] C. P. Rouke and B. J. Sanderson, Introduction to Piecewise-Linear Topology, Ergeb. der
Math. u. ihrer Grenz. 69 Springer (1972).
[15] H. Rubinstein and M. Scharlemann, Comparing Heegaard splittings of non-Haken 3-
manifolds, Topology 35 (1996), 1005-1026.
[16] M. Scharlemann and A. Thompson, Heegaard splittings of (surface)I are standard, Math.
Ann. 295 (1993), 549-564.
[17] M. Scharlemann and A. Thompson, Thin position and Heegaard splittings of the 3-sphere,
J. Dierential Geometry 295 (1993), 549-564.
[18] J. Schultens, The classication of Heegaard splittings for (compact orientable surfaces)S
1
,
Proc. London Math. Soc. 67 (1993), 425-448.
[19] E. Sedgwick, An innite collection of Heegaard splittings that are equivalent after one sta-
bilization, Math. Ann. 308 (1997), 65-72.
[20] J. Singer, Three-dimensional manifolds and their Heegaard diagrams, Trans. AMS 35 (1933),
88-111.
[21] F. Waldhausen, Heegaard-Zerlegungen der 3-sph are, Topology, 7 (1968), 195-203.
Department of Mathematics, Graduate School of Science, Osaka
University, Machikaneyama 1-16, Toyonaka, Osaka 560-0043, Japan
E-mail address: saito@gaia.math.wani.osaka-u.ac.jp
Mathematics Department, University of California, Santa Bar-
bara, CA93106, USA
E-mail address: mgscharl@math.ucsb.edu
Department of Mathematics, One Shields Avenue, University of
California, Davis, CA 95616, USA
E-mail address: jcs@math.ucdavis.edu

Toshio Saito, Martin Scharlemann and Jennifer Schultens - Lecture Notes On Generalized Heegaard Splittings

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Toshio Saito, Martin Scharlemann and Jennifer Schultens - Lecture Notes On Generalized Heegaard Splittings

Hochgeladen von

Copyright:

Verfügbare Formate

a

be the second barycentric subdivision of K. A regular

intersecting X (cf. Figure 3).

= X[0, 1], where X is identied with X{1/2} and (X; M) M is

C. A compression body C is called a handlebody if

C [0, 1] and (possibly empty) 3-balls. A complete meridian

C = , and cuts C into a 3-ball if

C [0, 1] and some (possibly empty) 3-balls. This gives a

C [0, 1] and some (possibly empty) 3-balls by attaching

C {1} and the boundary of the 3-balls (cf. Figure

be compression bodies. Suppose that C

C is an essential loop in the component

C. We shall prove that D A = . To this end, we

from A such that the interior of

is disjoint from D. Such a disk

. Then we obtain a disk D

and then attaching

(cf. Figure 6). Note that D

is a meridian disk of C. Moreover, we can isotope the interior of D

A| < |DA|, a contradiction. (Such an argument as above is called an

be a loop component of DA which is innermost in D, and let

cuts A into two annuli, and let A

be the component obtained by cutting A along

from A such that the interior of

is disjoint from D. Such a disk

is called an outermost disk for . Note that

(cf. Figure 7).

are inessential in C, then D is also inessential in C,

D A| < |D A|, a contradiction. Hence we have

is ambient isotopic to D if C is irreducible. Note that each component of

= . Take an ambient isotopy h

In the remainder of these notes, let M be a connected compact orientable 3-

be the dual 1-skeleton (see Figure 9).

is the barycentric subdivision of K. Note that no

C (cf. Figure 11).

= S (0 < t < 1). This is called a sweep-out picture.

) by adding a trivial handle. We say that (C

. Since genus(S) 2, we see that genus(S

; M) consists of two points, say p

; M) connecting the endpoint of

be a graph obtained from

is obtained from by an edge slide on .

. Particularly, the deformation of which is depicted as in Figure 15 is realized

= (; M). The notation h

, means a regular neighborhood of a vertex v of . Also, the

, means a 1-handle corresponding to an edge of

and a union of arcs and loops P in P. We choose P,

be an innermost disk for . Let

if P is a -reducing disk, or set P

if P is a reducing 2-sphere. If P is a -reducing disk, then P

|) < (|P |, |P |).

) whose interior is disjoint from

) such that the interior of

(cf. Figure 16).

is a minimal complete meridian system of C

| < |P |. This contradicts the minimality of

which is a meridian disk of C

D be a disk obtained by cutting D along

an outermost disk for (D

may intersects P transversely (cf.

to reduce |P | (cf. Figure 22).

be the endpoints of . Let and